xref: /dragonfly/sys/vfs/nfs/nfs_vnops.c (revision ae071d8d)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
33  * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
34  */
35 
36 
37 /*
38  * vnode op calls for Sun NFS version 2 and 3
39  */
40 
41 #include "opt_inet.h"
42 
43 #include <sys/param.h>
44 #include <sys/kernel.h>
45 #include <sys/systm.h>
46 #include <sys/resourcevar.h>
47 #include <sys/proc.h>
48 #include <sys/mount.h>
49 #include <sys/buf.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/namei.h>
53 #include <sys/nlookup.h>
54 #include <sys/socket.h>
55 #include <sys/vnode.h>
56 #include <sys/dirent.h>
57 #include <sys/fcntl.h>
58 #include <sys/lockf.h>
59 #include <sys/stat.h>
60 #include <sys/sysctl.h>
61 #include <sys/conf.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_extern.h>
65 
66 #include <sys/buf2.h>
67 
68 #include <vfs/fifofs/fifo.h>
69 #include <vfs/ufs/dir.h>
70 
71 #undef DIRBLKSIZ
72 
73 #include "rpcv2.h"
74 #include "nfsproto.h"
75 #include "nfs.h"
76 #include "nfsmount.h"
77 #include "nfsnode.h"
78 #include "xdr_subs.h"
79 #include "nfsm_subs.h"
80 
81 #include <net/if.h>
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
84 
85 #include <sys/thread2.h>
86 
87 /* Defs */
88 #define	TRUE	1
89 #define	FALSE	0
90 
91 static int	nfsfifo_read (struct vop_read_args *);
92 static int	nfsfifo_write (struct vop_write_args *);
93 static int	nfsfifo_close (struct vop_close_args *);
94 static int	nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
95 static	int	nfs_lookup (struct vop_old_lookup_args *);
96 static	int	nfs_create (struct vop_old_create_args *);
97 static	int	nfs_mknod (struct vop_old_mknod_args *);
98 static	int	nfs_open (struct vop_open_args *);
99 static	int	nfs_close (struct vop_close_args *);
100 static	int	nfs_access (struct vop_access_args *);
101 static	int	nfs_getattr (struct vop_getattr_args *);
102 static	int	nfs_setattr (struct vop_setattr_args *);
103 static	int	nfs_read (struct vop_read_args *);
104 static	int	nfs_mmap (struct vop_mmap_args *);
105 static	int	nfs_fsync (struct vop_fsync_args *);
106 static	int	nfs_remove (struct vop_old_remove_args *);
107 static	int	nfs_link (struct vop_old_link_args *);
108 static	int	nfs_rename (struct vop_old_rename_args *);
109 static	int	nfs_mkdir (struct vop_old_mkdir_args *);
110 static	int	nfs_rmdir (struct vop_old_rmdir_args *);
111 static	int	nfs_symlink (struct vop_old_symlink_args *);
112 static	int	nfs_readdir (struct vop_readdir_args *);
113 static	int	nfs_bmap (struct vop_bmap_args *);
114 static	int	nfs_strategy (struct vop_strategy_args *);
115 static	int	nfs_lookitup (struct vnode *, const char *, int,
116 			struct ucred *, struct thread *, struct nfsnode **);
117 static	int	nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
118 static int	nfs_laccess (struct vop_access_args *);
119 static int	nfs_readlink (struct vop_readlink_args *);
120 static int	nfs_print (struct vop_print_args *);
121 static int	nfs_advlock (struct vop_advlock_args *);
122 
123 static	int	nfs_nresolve (struct vop_nresolve_args *);
124 /*
125  * Global vfs data structures for nfs
126  */
127 struct vop_ops nfsv2_vnode_vops = {
128 	.vop_default =		vop_defaultop,
129 	.vop_access =		nfs_access,
130 	.vop_advlock =		nfs_advlock,
131 	.vop_bmap =		nfs_bmap,
132 	.vop_close =		nfs_close,
133 	.vop_old_create =	nfs_create,
134 	.vop_fsync =		nfs_fsync,
135 	.vop_getattr =		nfs_getattr,
136 	.vop_getpages =		vop_stdgetpages,
137 	.vop_putpages =		vop_stdputpages,
138 	.vop_inactive =		nfs_inactive,
139 	.vop_old_link =		nfs_link,
140 	.vop_old_lookup =	nfs_lookup,
141 	.vop_old_mkdir =	nfs_mkdir,
142 	.vop_old_mknod =	nfs_mknod,
143 	.vop_mmap =		nfs_mmap,
144 	.vop_open =		nfs_open,
145 	.vop_print =		nfs_print,
146 	.vop_read =		nfs_read,
147 	.vop_readdir =		nfs_readdir,
148 	.vop_readlink =		nfs_readlink,
149 	.vop_reclaim =		nfs_reclaim,
150 	.vop_old_remove =	nfs_remove,
151 	.vop_old_rename =	nfs_rename,
152 	.vop_old_rmdir =	nfs_rmdir,
153 	.vop_setattr =		nfs_setattr,
154 	.vop_strategy =		nfs_strategy,
155 	.vop_old_symlink =	nfs_symlink,
156 	.vop_write =		nfs_write,
157 	.vop_nresolve =		nfs_nresolve
158 };
159 
160 /*
161  * Special device vnode ops
162  */
163 struct vop_ops nfsv2_spec_vops = {
164 	.vop_default =		vop_defaultop,
165 	.vop_access =		nfs_laccess,
166 	.vop_close =		nfs_close,
167 	.vop_fsync =		nfs_fsync,
168 	.vop_getattr =		nfs_getattr,
169 	.vop_inactive =		nfs_inactive,
170 	.vop_print =		nfs_print,
171 	.vop_read =		vop_stdnoread,
172 	.vop_reclaim =		nfs_reclaim,
173 	.vop_setattr =		nfs_setattr,
174 	.vop_write =		vop_stdnowrite
175 };
176 
177 struct vop_ops nfsv2_fifo_vops = {
178 	.vop_default =		fifo_vnoperate,
179 	.vop_access =		nfs_laccess,
180 	.vop_close =		nfsfifo_close,
181 	.vop_fsync =		nfs_fsync,
182 	.vop_getattr =		nfs_getattr,
183 	.vop_inactive =		nfs_inactive,
184 	.vop_print =		nfs_print,
185 	.vop_read =		nfsfifo_read,
186 	.vop_reclaim =		nfs_reclaim,
187 	.vop_setattr =		nfs_setattr,
188 	.vop_write =		nfsfifo_write
189 };
190 
191 static int	nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
192 				  struct componentname *cnp,
193 				  struct vattr *vap);
194 static int	nfs_removerpc (struct vnode *dvp, const char *name,
195 				   int namelen,
196 				   struct ucred *cred, struct thread *td);
197 static int	nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
198 				   int fnamelen, struct vnode *tdvp,
199 				   const char *tnameptr, int tnamelen,
200 				   struct ucred *cred, struct thread *td);
201 static int	nfs_renameit (struct vnode *sdvp,
202 				  struct componentname *scnp,
203 				  struct sillyrename *sp);
204 
205 SYSCTL_DECL(_vfs_nfs);
206 
207 static int nfs_flush_on_rename = 1;
208 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
209 	   &nfs_flush_on_rename, 0, "flush fvp prior to rename");
210 static int nfs_flush_on_hlink = 0;
211 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
212 	   &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
213 
214 static int	nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
215 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
216 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
217 
218 static int	nfsneg_cache_timeout = NFS_MINATTRTIMO;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
220 	   &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
221 
222 static int	nfspos_cache_timeout = NFS_MINATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
224 	   &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
225 
226 static int	nfsv3_commit_on_close = 0;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
228 	   &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
229 #if 0
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
231 	   &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
232 
233 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
234 	   &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
235 #endif
236 
237 #define	NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY		\
238 			 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE	\
239 			 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
240 
241 /*
242  * Returns whether a name component is a degenerate '.' or '..'.
243  */
244 static __inline
245 int
246 nlcdegenerate(struct nlcomponent *nlc)
247 {
248 	if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
249 		return(1);
250 	if (nlc->nlc_namelen == 2 &&
251 	    nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
252 		return(1);
253 	return(0);
254 }
255 
256 static int
257 nfs3_access_otw(struct vnode *vp, int wmode,
258 		struct thread *td, struct ucred *cred)
259 {
260 	struct nfsnode *np = VTONFS(vp);
261 	int attrflag;
262 	int error = 0;
263 	u_int32_t *tl;
264 	u_int32_t rmode;
265 	struct nfsm_info info;
266 
267 	info.mrep = NULL;
268 	info.v3 = 1;
269 
270 	nfsstats.rpccnt[NFSPROC_ACCESS]++;
271 	nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
272 		     NFSX_FH(info.v3) + NFSX_UNSIGNED);
273 	ERROROUT(nfsm_fhtom(&info, vp));
274 	tl = nfsm_build(&info, NFSX_UNSIGNED);
275 	*tl = txdr_unsigned(wmode);
276 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
277 	ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
278 	if (error == 0) {
279 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
280 		rmode = fxdr_unsigned(u_int32_t, *tl);
281 		np->n_mode = rmode;
282 		np->n_modeuid = cred->cr_uid;
283 		np->n_modestamp = mycpu->gd_time_seconds;
284 	}
285 	m_freem(info.mrep);
286 	info.mrep = NULL;
287 nfsmout:
288 	return error;
289 }
290 
291 /*
292  * nfs access vnode op.
293  * For nfs version 2, just return ok. File accesses may fail later.
294  * For nfs version 3, use the access rpc to check accessibility. If file modes
295  * are changed on the server, accesses might still fail later.
296  *
297  * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
298  */
299 static int
300 nfs_access(struct vop_access_args *ap)
301 {
302 	struct ucred *cred;
303 	struct vnode *vp = ap->a_vp;
304 	thread_t td = curthread;
305 	int error = 0;
306 	u_int32_t mode, wmode;
307 	struct nfsnode *np = VTONFS(vp);
308 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
309 	int v3 = NFS_ISV3(vp);
310 
311 	lwkt_gettoken(&nmp->nm_token);
312 
313 	/*
314 	 * Disallow write attempts on filesystems mounted read-only;
315 	 * unless the file is a socket, fifo, or a block or character
316 	 * device resident on the filesystem.
317 	 */
318 	if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
319 		switch (vp->v_type) {
320 		case VREG:
321 		case VDIR:
322 		case VLNK:
323 			lwkt_reltoken(&nmp->nm_token);
324 			return (EROFS);
325 		default:
326 			break;
327 		}
328 	}
329 
330 	/*
331 	 * The NFS protocol passes only the effective uid/gid over the wire but
332 	 * we need to check access against real ids if AT_EACCESS not set.
333 	 * Handle this case by cloning the credentials and setting the
334 	 * effective ids to the real ones.
335 	 */
336 	if (ap->a_flags & AT_EACCESS) {
337 		cred = crhold(ap->a_cred);
338 	} else {
339 		cred = crdup(ap->a_cred);
340 		cred->cr_uid = cred->cr_ruid;
341 		cred->cr_gid = cred->cr_rgid;
342 	}
343 
344 	/*
345 	 * For nfs v3, check to see if we have done this recently, and if
346 	 * so return our cached result instead of making an ACCESS call.
347 	 * If not, do an access rpc, otherwise you are stuck emulating
348 	 * ufs_access() locally using the vattr. This may not be correct,
349 	 * since the server may apply other access criteria such as
350 	 * client uid-->server uid mapping that we do not know about.
351 	 */
352 	if (v3) {
353 		if (ap->a_mode & VREAD)
354 			mode = NFSV3ACCESS_READ;
355 		else
356 			mode = 0;
357 		if (vp->v_type != VDIR) {
358 			if (ap->a_mode & VWRITE)
359 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
360 			if (ap->a_mode & VEXEC)
361 				mode |= NFSV3ACCESS_EXECUTE;
362 		} else {
363 			if (ap->a_mode & VWRITE)
364 				mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
365 					 NFSV3ACCESS_DELETE);
366 			if (ap->a_mode & VEXEC)
367 				mode |= NFSV3ACCESS_LOOKUP;
368 		}
369 		/* XXX safety belt, only make blanket request if caching */
370 		if (nfsaccess_cache_timeout > 0) {
371 			wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
372 				NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
373 				NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
374 		} else {
375 			wmode = mode;
376 		}
377 
378 		/*
379 		 * Does our cached result allow us to give a definite yes to
380 		 * this request?
381 		 */
382 		if (np->n_modestamp &&
383 		   (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
384 		   (cred->cr_uid == np->n_modeuid) &&
385 		   ((np->n_mode & mode) == mode)) {
386 			nfsstats.accesscache_hits++;
387 		} else {
388 			/*
389 			 * Either a no, or a don't know.  Go to the wire.
390 			 */
391 			nfsstats.accesscache_misses++;
392 		        error = nfs3_access_otw(vp, wmode, td, cred);
393 			if (!error) {
394 				if ((np->n_mode & mode) != mode) {
395 					error = EACCES;
396 				}
397 			}
398 		}
399 	} else {
400 		if ((error = nfs_laccess(ap)) != 0) {
401 			crfree(cred);
402 			lwkt_reltoken(&nmp->nm_token);
403 			return (error);
404 		}
405 
406 		/*
407 		 * Attempt to prevent a mapped root from accessing a file
408 		 * which it shouldn't.  We try to read a byte from the file
409 		 * if the user is root and the file is not zero length.
410 		 * After calling nfs_laccess, we should have the correct
411 		 * file size cached.
412 		 */
413 		if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
414 		    && VTONFS(vp)->n_size > 0) {
415 			struct iovec aiov;
416 			struct uio auio;
417 			char buf[1];
418 
419 			aiov.iov_base = buf;
420 			aiov.iov_len = 1;
421 			auio.uio_iov = &aiov;
422 			auio.uio_iovcnt = 1;
423 			auio.uio_offset = 0;
424 			auio.uio_resid = 1;
425 			auio.uio_segflg = UIO_SYSSPACE;
426 			auio.uio_rw = UIO_READ;
427 			auio.uio_td = td;
428 
429 			if (vp->v_type == VREG) {
430 				error = nfs_readrpc_uio(vp, &auio);
431 			} else if (vp->v_type == VDIR) {
432 				char* bp;
433 				bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
434 				aiov.iov_base = bp;
435 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
436 				error = nfs_readdirrpc_uio(vp, &auio);
437 				kfree(bp, M_TEMP);
438 			} else if (vp->v_type == VLNK) {
439 				error = nfs_readlinkrpc_uio(vp, &auio);
440 			} else {
441 				error = EACCES;
442 			}
443 		}
444 	}
445 	/*
446 	 * [re]record creds for reading and/or writing if access
447 	 * was granted.  Assume the NFS server will grant read access
448 	 * for execute requests.
449 	 */
450 	if (error == 0) {
451 		if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
452 			crhold(cred);
453 			if (np->n_rucred)
454 				crfree(np->n_rucred);
455 			np->n_rucred = cred;
456 		}
457 		if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
458 			crhold(cred);
459 			if (np->n_wucred)
460 				crfree(np->n_wucred);
461 			np->n_wucred = cred;
462 		}
463 	}
464 	lwkt_reltoken(&nmp->nm_token);
465 	crfree(cred);
466 	return(error);
467 }
468 
469 /*
470  * nfs open vnode op
471  * Check to see if the type is ok
472  * and that deletion is not in progress.
473  * For paged in text files, you will need to flush the page cache
474  * if consistency is lost.
475  *
476  * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
477  *	    struct file *a_fp)
478  */
479 /* ARGSUSED */
480 static int
481 nfs_open(struct vop_open_args *ap)
482 {
483 	struct vnode *vp = ap->a_vp;
484 	struct nfsnode *np = VTONFS(vp);
485 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
486 	struct vattr vattr;
487 	int error;
488 
489 	lwkt_gettoken(&nmp->nm_token);
490 
491 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
492 #ifdef DIAGNOSTIC
493 		kprintf("open eacces vtyp=%d\n",vp->v_type);
494 #endif
495 		lwkt_reltoken(&nmp->nm_token);
496 		return (EOPNOTSUPP);
497 	}
498 
499 	/*
500 	 * Save valid creds for reading and writing for later RPCs.
501 	 */
502 	if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
503 		crhold(ap->a_cred);
504 		if (np->n_rucred)
505 			crfree(np->n_rucred);
506 		np->n_rucred = ap->a_cred;
507 	}
508 	if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
509 		crhold(ap->a_cred);
510 		if (np->n_wucred)
511 			crfree(np->n_wucred);
512 		np->n_wucred = ap->a_cred;
513 	}
514 
515 	/*
516 	 * Clear the attribute cache only if opening with write access.  It
517 	 * is unclear if we should do this at all here, but we certainly
518 	 * should not clear the cache unconditionally simply because a file
519 	 * is being opened.
520 	 */
521 	if (ap->a_mode & FWRITE)
522 		np->n_attrstamp = 0;
523 
524 	/*
525 	 * For normal NFS, reconcile changes made locally verses
526 	 * changes made remotely.  Note that VOP_GETATTR only goes
527 	 * to the wire if the cached attribute has timed out or been
528 	 * cleared.
529 	 *
530 	 * If local modifications have been made clear the attribute
531 	 * cache to force an attribute and modified time check.  If
532 	 * GETATTR detects that the file has been changed by someone
533 	 * other then us it will set NRMODIFIED.
534 	 *
535 	 * If we are opening a directory and local changes have been
536 	 * made we have to invalidate the cache in order to ensure
537 	 * that we get the most up-to-date information from the
538 	 * server.  XXX
539 	 */
540 	if (np->n_flag & NLMODIFIED) {
541 		np->n_attrstamp = 0;
542 		if (vp->v_type == VDIR) {
543 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
544 			if (error == EINTR)
545 				return (error);
546 			nfs_invaldir(vp);
547 		}
548 	}
549 	error = VOP_GETATTR(vp, &vattr);
550 	if (error) {
551 		lwkt_reltoken(&nmp->nm_token);
552 		return (error);
553 	}
554 	if (np->n_flag & NRMODIFIED) {
555 		if (vp->v_type == VDIR)
556 			nfs_invaldir(vp);
557 		error = nfs_vinvalbuf(vp, V_SAVE, 1);
558 		if (error == EINTR) {
559 			lwkt_reltoken(&nmp->nm_token);
560 			return (error);
561 		}
562 		np->n_flag &= ~NRMODIFIED;
563 	}
564 	error = vop_stdopen(ap);
565 	lwkt_reltoken(&nmp->nm_token);
566 
567 	return error;
568 }
569 
570 /*
571  * nfs close vnode op
572  * What an NFS client should do upon close after writing is a debatable issue.
573  * Most NFS clients push delayed writes to the server upon close, basically for
574  * two reasons:
575  * 1 - So that any write errors may be reported back to the client process
576  *     doing the close system call. By far the two most likely errors are
577  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
578  * 2 - To put a worst case upper bound on cache inconsistency between
579  *     multiple clients for the file.
580  * There is also a consistency problem for Version 2 of the protocol w.r.t.
581  * not being able to tell if other clients are writing a file concurrently,
582  * since there is no way of knowing if the changed modify time in the reply
583  * is only due to the write for this client.
584  * (NFS Version 3 provides weak cache consistency data in the reply that
585  *  should be sufficient to detect and handle this case.)
586  *
587  * The current code does the following:
588  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
589  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
590  *                     or commit them (this satisfies 1 and 2 except for the
591  *                     case where the server crashes after this close but
592  *                     before the commit RPC, which is felt to be "good
593  *                     enough". Changing the last argument to nfs_flush() to
594  *                     a 1 would force a commit operation, if it is felt a
595  *                     commit is necessary now.
596  * for NQNFS         - do nothing now, since 2 is dealt with via leases and
597  *                     1 should be dealt with via an fsync() system call for
598  *                     cases where write errors are important.
599  *
600  * nfs_close(struct vnode *a_vp, int a_fflag)
601  */
602 /* ARGSUSED */
603 static int
604 nfs_close(struct vop_close_args *ap)
605 {
606 	struct vnode *vp = ap->a_vp;
607 	struct nfsnode *np = VTONFS(vp);
608 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
609 	int error = 0;
610 	thread_t td = curthread;
611 
612 	vn_lock(vp, LK_UPGRADE | LK_RETRY); /* XXX */
613 	lwkt_gettoken(&nmp->nm_token);
614 
615 	if (vp->v_type == VREG) {
616 	    if (np->n_flag & NLMODIFIED) {
617 		if (NFS_ISV3(vp)) {
618 		    /*
619 		     * Under NFSv3 we have dirty buffers to dispose of.  We
620 		     * must flush them to the NFS server.  We have the option
621 		     * of waiting all the way through the commit rpc or just
622 		     * waiting for the initial write.  The default is to only
623 		     * wait through the initial write so the data is in the
624 		     * server's cache, which is roughly similar to the state
625 		     * a standard disk subsystem leaves the file in on close().
626 		     *
627 		     * We cannot clear the NLMODIFIED bit in np->n_flag due to
628 		     * potential races with other processes, and certainly
629 		     * cannot clear it if we don't commit.
630 		     */
631 		    int cm = nfsv3_commit_on_close ? 1 : 0;
632 		    error = nfs_flush(vp, MNT_WAIT, td, cm);
633 		    /* np->n_flag &= ~NLMODIFIED; */
634 		} else {
635 		    error = nfs_vinvalbuf(vp, V_SAVE, 1);
636 		}
637 		np->n_attrstamp = 0;
638 	    }
639 	    if (np->n_flag & NWRITEERR) {
640 		np->n_flag &= ~NWRITEERR;
641 		error = np->n_error;
642 	    }
643 	}
644 	vop_stdclose(ap);
645 	lwkt_reltoken(&nmp->nm_token);
646 
647 	return (error);
648 }
649 
650 /*
651  * nfs getattr call from vfs.
652  *
653  * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
654  */
655 static int
656 nfs_getattr(struct vop_getattr_args *ap)
657 {
658 	struct vnode *vp = ap->a_vp;
659 	struct nfsnode *np = VTONFS(vp);
660 	struct nfsmount *nmp;
661 	int error = 0;
662 	thread_t td = curthread;
663 	struct nfsm_info info;
664 
665 	info.mrep = NULL;
666 	info.v3 = NFS_ISV3(vp);
667 	nmp = VFSTONFS(vp->v_mount);
668 
669 	lwkt_gettoken(&nmp->nm_token);
670 
671 	/*
672 	 * Update local times for special files.
673 	 */
674 	if (np->n_flag & (NACC | NUPD))
675 		np->n_flag |= NCHG;
676 	/*
677 	 * First look in the cache.
678 	 */
679 	if (nfs_getattrcache(vp, ap->a_vap) == 0)
680 		goto done;
681 
682 	if (info.v3 && nfsaccess_cache_timeout > 0) {
683 		nfsstats.accesscache_misses++;
684 		nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
685 		if (nfs_getattrcache(vp, ap->a_vap) == 0)
686 			goto done;
687 	}
688 
689 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
690 	nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
691 	ERROROUT(nfsm_fhtom(&info, vp));
692 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
693 				nfs_vpcred(vp, ND_CHECK), &error));
694 	if (error == 0) {
695 		ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
696 	}
697 	m_freem(info.mrep);
698 	info.mrep = NULL;
699 done:
700 	/*
701 	 * NFS doesn't support chflags flags.  If the nfs mount was
702 	 * made -o cache set the UF_CACHE bit for swapcache.
703 	 */
704 	if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
705 		ap->a_vap->va_flags |= UF_CACHE;
706 nfsmout:
707 	lwkt_reltoken(&nmp->nm_token);
708 	return (error);
709 }
710 
711 /*
712  * nfs setattr call.
713  *
714  * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
715  */
716 static int
717 nfs_setattr(struct vop_setattr_args *ap)
718 {
719 	struct vnode *vp = ap->a_vp;
720 	struct nfsnode *np = VTONFS(vp);
721 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
722 	struct vattr *vap = ap->a_vap;
723 	int error = 0;
724 	off_t tsize;
725 	thread_t td = curthread;
726 
727 #ifndef nolint
728 	tsize = (off_t)0;
729 #endif
730 	/*
731 	 * Setting of flags is not supported.
732 	 */
733 	if (vap->va_flags != VNOVAL)
734 		return (EOPNOTSUPP);
735 
736 	/*
737 	 * Disallow write attempts if the filesystem is mounted read-only.
738 	 */
739   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
740 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
741 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
742 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
743 		return (EROFS);
744 
745 	lwkt_gettoken(&nmp->nm_token);
746 
747 	if (vap->va_size != VNOVAL) {
748 		/*
749 		 * truncation requested
750 		 */
751  		switch (vp->v_type) {
752  		case VDIR:
753 			lwkt_reltoken(&nmp->nm_token);
754  			return (EISDIR);
755  		case VCHR:
756  		case VBLK:
757  		case VSOCK:
758  		case VFIFO:
759 			if (vap->va_mtime.tv_sec == VNOVAL &&
760 			    vap->va_atime.tv_sec == VNOVAL &&
761 			    vap->va_mode == (mode_t)VNOVAL &&
762 			    vap->va_uid == (uid_t)VNOVAL &&
763 			    vap->va_gid == (gid_t)VNOVAL) {
764 				lwkt_reltoken(&nmp->nm_token);
765 				return (0);
766 			}
767  			vap->va_size = VNOVAL;
768  			break;
769  		default:
770 			/*
771 			 * Disallow write attempts if the filesystem is
772 			 * mounted read-only.
773 			 */
774 			if (vp->v_mount->mnt_flag & MNT_RDONLY) {
775 				lwkt_reltoken(&nmp->nm_token);
776 				return (EROFS);
777 			}
778 
779 			tsize = np->n_size;
780 again:
781 			error = nfs_meta_setsize(vp, td, vap->va_size, 0);
782 
783 #if 0
784  			if (np->n_flag & NLMODIFIED) {
785  			    if (vap->va_size == 0)
786  				error = nfs_vinvalbuf(vp, 0, 1);
787  			    else
788  				error = nfs_vinvalbuf(vp, V_SAVE, 1);
789  			}
790 #endif
791 			/*
792 			 * note: this loop case almost always happens at
793 			 * least once per truncation.
794 			 */
795 			if (error == 0 && np->n_size != vap->va_size)
796 				goto again;
797 			np->n_vattr.va_size = vap->va_size;
798 			break;
799 		}
800 	} else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
801 		/*
802 		 * What to do.  If we are modifying the mtime we lose
803 		 * mtime detection of changes made by the server or other
804 		 * clients.  But programs like rsync/rdist/cpdup are going
805 		 * to call utimes a lot.  We don't want to piecemeal sync.
806 		 *
807 		 * For now sync if any prior remote changes were detected,
808 		 * but allow us to lose track of remote changes made during
809 		 * the utimes operation.
810 		 */
811 		if (np->n_flag & NRMODIFIED)
812 			error = nfs_vinvalbuf(vp, V_SAVE, 1);
813 		if (error == EINTR)
814 			return (error);
815 		if (error == 0) {
816 			if (vap->va_mtime.tv_sec != VNOVAL) {
817 				np->n_mtime = vap->va_mtime.tv_sec;
818 			}
819 		}
820 	}
821 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
822 
823 	/*
824 	 * Sanity check if a truncation was issued.  This should only occur
825 	 * if multiple processes are racing on the same file.
826 	 */
827 	if (error == 0 && vap->va_size != VNOVAL &&
828 	    np->n_size != vap->va_size) {
829 		kprintf("NFS ftruncate: server disagrees on the file size: "
830 			"%jd/%jd/%jd\n",
831 			(intmax_t)tsize,
832 			(intmax_t)vap->va_size,
833 			(intmax_t)np->n_size);
834 		goto again;
835 	}
836 	if (error && vap->va_size != VNOVAL) {
837 		np->n_size = np->n_vattr.va_size = tsize;
838 		nfs_meta_setsize(vp, td, np->n_size, 0);
839 	}
840 	lwkt_reltoken(&nmp->nm_token);
841 
842 	return (error);
843 }
844 
845 /*
846  * Do an nfs setattr rpc.
847  */
848 static int
849 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
850 	       struct ucred *cred, struct thread *td)
851 {
852 	struct nfsv2_sattr *sp;
853 	struct nfsnode *np = VTONFS(vp);
854 	u_int32_t *tl;
855 	int error = 0, wccflag = NFSV3_WCCRATTR;
856 	struct nfsm_info info;
857 
858 	info.mrep = NULL;
859 	info.v3 = NFS_ISV3(vp);
860 
861 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
862 	nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
863 		     NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
864 	ERROROUT(nfsm_fhtom(&info, vp));
865 	if (info.v3) {
866 		nfsm_v3attrbuild(&info, vap, TRUE);
867 		tl = nfsm_build(&info, NFSX_UNSIGNED);
868 		*tl = nfs_false;
869 	} else {
870 		sp = nfsm_build(&info, NFSX_V2SATTR);
871 		if (vap->va_mode == (mode_t)VNOVAL)
872 			sp->sa_mode = nfs_xdrneg1;
873 		else
874 			sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
875 		if (vap->va_uid == (uid_t)VNOVAL)
876 			sp->sa_uid = nfs_xdrneg1;
877 		else
878 			sp->sa_uid = txdr_unsigned(vap->va_uid);
879 		if (vap->va_gid == (gid_t)VNOVAL)
880 			sp->sa_gid = nfs_xdrneg1;
881 		else
882 			sp->sa_gid = txdr_unsigned(vap->va_gid);
883 		sp->sa_size = txdr_unsigned(vap->va_size);
884 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
885 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
886 	}
887 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
888 	if (info.v3) {
889 		np->n_modestamp = 0;
890 		ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
891 	} else {
892 		ERROROUT(nfsm_loadattr(&info, vp, NULL));
893 	}
894 	m_freem(info.mrep);
895 	info.mrep = NULL;
896 nfsmout:
897 	return (error);
898 }
899 
900 static
901 void
902 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
903 {
904 	if (nctimeout == 0)
905 		nctimeout = 1;
906 	else
907 		nctimeout *= hz;
908 	cache_setvp(nch, vp);
909 	cache_settimeout(nch, nctimeout);
910 }
911 
912 /*
913  * NEW API CALL - replaces nfs_lookup().  However, we cannot remove
914  * nfs_lookup() until all remaining new api calls are implemented.
915  *
916  * Resolve a namecache entry.  This function is passed a locked ncp and
917  * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
918  */
919 static int
920 nfs_nresolve(struct vop_nresolve_args *ap)
921 {
922 	struct thread *td = curthread;
923 	struct namecache *ncp;
924 	struct nfsmount *nmp;
925 	struct nfsnode *np;
926 	struct vnode *dvp;
927 	struct vnode *nvp;
928 	nfsfh_t *fhp;
929 	int attrflag;
930 	int fhsize;
931 	int error;
932 	int tmp_error;
933 	int len;
934 	struct nfsm_info info;
935 
936 	dvp = ap->a_dvp;
937 	nmp = VFSTONFS(dvp->v_mount);
938 
939 	lwkt_gettoken(&nmp->nm_token);
940 
941 	if ((error = vget(dvp, LK_SHARED)) != 0) {
942 		lwkt_reltoken(&nmp->nm_token);
943 		return (error);
944 	}
945 
946 	info.mrep = NULL;
947 	info.v3 = NFS_ISV3(dvp);
948 
949 	nvp = NULL;
950 	nfsstats.lookupcache_misses++;
951 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
952 	ncp = ap->a_nch->ncp;
953 	len = ncp->nc_nlen;
954 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
955 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
956 	ERROROUT(nfsm_fhtom(&info, dvp));
957 	ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
958 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
959 				ap->a_cred, &error));
960 	if (error) {
961 		/*
962 		 * Cache negatve lookups to reduce NFS traffic, but use
963 		 * a fast timeout.  Otherwise use a timeout of 1 tick.
964 		 * XXX we should add a namecache flag for no-caching
965 		 * to uncache the negative hit as soon as possible, but
966 		 * we cannot simply destroy the entry because it is used
967 		 * as a placeholder by the caller.
968 		 *
969 		 * The refactored nfs code will overwrite a non-zero error
970 		 * with 0 when we use ERROROUT(), so don't here.
971 		 */
972 		if (error == ENOENT)
973 			nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
974 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
975 					     NFS_LATTR_NOSHRINK);
976 		if (tmp_error) {
977 			error = tmp_error;
978 			goto nfsmout;
979 		}
980 		m_freem(info.mrep);
981 		info.mrep = NULL;
982 		goto nfsmout;
983 	}
984 
985 	/*
986 	 * Success, get the file handle, do various checks, and load
987 	 * post-operation data from the reply packet.  Theoretically
988 	 * we should never be looking up "." so, theoretically, we
989 	 * should never get the same file handle as our directory.  But
990 	 * we check anyway. XXX
991 	 *
992 	 * Note that no timeout is set for the positive cache hit.  We
993 	 * assume, theoretically, that ESTALE returns will be dealt with
994 	 * properly to handle NFS races and in anycase we cannot depend
995 	 * on a timeout to deal with NFS open/create/excl issues so instead
996 	 * of a bad hack here the rest of the NFS client code needs to do
997 	 * the right thing.
998 	 */
999 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1000 
1001 	np = VTONFS(dvp);
1002 	if (NFS_CMPFH(np, fhp, fhsize)) {
1003 		vref(dvp);
1004 		nvp = dvp;
1005 	} else {
1006 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, NULL);
1007 		if (error) {
1008 			m_freem(info.mrep);
1009 			info.mrep = NULL;
1010 			vput(dvp);
1011 			lwkt_reltoken(&nmp->nm_token);
1012 			return (error);
1013 		}
1014 		nvp = NFSTOV(np);
1015 	}
1016 	if (info.v3) {
1017 		ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
1018 					  NFS_LATTR_NOSHRINK));
1019 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1020 					  NFS_LATTR_NOSHRINK));
1021 	} else {
1022 		ERROROUT(nfsm_loadattr(&info, nvp, NULL));
1023 	}
1024 	nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1025 	m_freem(info.mrep);
1026 	info.mrep = NULL;
1027 nfsmout:
1028 	lwkt_reltoken(&nmp->nm_token);
1029 	vput(dvp);
1030 	if (nvp) {
1031 		if (nvp == dvp)
1032 			vrele(nvp);
1033 		else
1034 			vput(nvp);
1035 	}
1036 	return (error);
1037 }
1038 
1039 /*
1040  * 'cached' nfs directory lookup
1041  *
1042  * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1043  *
1044  * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1045  *	      struct componentname *a_cnp)
1046  */
1047 static int
1048 nfs_lookup(struct vop_old_lookup_args *ap)
1049 {
1050 	struct componentname *cnp = ap->a_cnp;
1051 	struct vnode *dvp = ap->a_dvp;
1052 	struct vnode **vpp = ap->a_vpp;
1053 	int flags = cnp->cn_flags;
1054 	struct vnode *newvp;
1055 	struct vnode *notvp;
1056 	struct nfsmount *nmp;
1057 	long len;
1058 	nfsfh_t *fhp;
1059 	struct nfsnode *np;
1060 	int lockparent, wantparent, attrflag, fhsize;
1061 	int error;
1062 	int tmp_error;
1063 	struct nfsm_info info;
1064 
1065 	info.mrep = NULL;
1066 	info.v3 = NFS_ISV3(dvp);
1067 	error = 0;
1068 
1069 	notvp = (cnp->cn_flags & CNP_NOTVP) ? cnp->cn_notvp : NULL;
1070 
1071 	/*
1072 	 * Read-only mount check and directory check.
1073 	 */
1074 	*vpp = NULLVP;
1075 	if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1076 	    (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1077 		return (EROFS);
1078 
1079 	if (dvp->v_type != VDIR)
1080 		return (ENOTDIR);
1081 
1082 	/*
1083 	 * Look it up in the cache.  Note that ENOENT is only returned if we
1084 	 * previously entered a negative hit (see later on).  The additional
1085 	 * nfsneg_cache_timeout check causes previously cached results to
1086 	 * be instantly ignored if the negative caching is turned off.
1087 	 */
1088 	lockparent = flags & CNP_LOCKPARENT;
1089 	wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1090 	nmp = VFSTONFS(dvp->v_mount);
1091 	np = VTONFS(dvp);
1092 
1093 	lwkt_gettoken(&nmp->nm_token);
1094 
1095 	/*
1096 	 * Go to the wire.
1097 	 */
1098 	error = 0;
1099 	newvp = NULLVP;
1100 	nfsstats.lookupcache_misses++;
1101 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1102 	len = cnp->cn_namelen;
1103 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1104 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1105 	ERROROUT(nfsm_fhtom(&info, dvp));
1106 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1107 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1108 				cnp->cn_cred, &error));
1109 	if (error) {
1110 		tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1111 					     NFS_LATTR_NOSHRINK);
1112 		if (tmp_error) {
1113 			error = tmp_error;
1114 			goto nfsmout;
1115 		}
1116 
1117 		m_freem(info.mrep);
1118 		info.mrep = NULL;
1119 		goto nfsmout;
1120 	}
1121 	NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1122 
1123 	/*
1124 	 * Handle RENAME case...
1125 	 */
1126 	if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1127 		if (NFS_CMPFH(np, fhp, fhsize)) {
1128 			m_freem(info.mrep);
1129 			info.mrep = NULL;
1130 			lwkt_reltoken(&nmp->nm_token);
1131 			return (EISDIR);
1132 		}
1133 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1134 		if (error) {
1135 			m_freem(info.mrep);
1136 			info.mrep = NULL;
1137 			lwkt_reltoken(&nmp->nm_token);
1138 			return (error);
1139 		}
1140 		newvp = NFSTOV(np);
1141 		if (info.v3) {
1142 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1143 						  NFS_LATTR_NOSHRINK));
1144 			ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1145 						  NFS_LATTR_NOSHRINK));
1146 		} else {
1147 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1148 		}
1149 		*vpp = newvp;
1150 		m_freem(info.mrep);
1151 		info.mrep = NULL;
1152 		if (!lockparent) {
1153 			vn_unlock(dvp);
1154 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1155 		}
1156 		lwkt_reltoken(&nmp->nm_token);
1157 		return (0);
1158 	}
1159 
1160 	if (flags & CNP_ISDOTDOT) {
1161 		vn_unlock(dvp);
1162 		cnp->cn_flags |= CNP_PDIRUNLOCK;
1163 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1164 		if (error) {
1165 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1166 			cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1167 			lwkt_reltoken(&nmp->nm_token);
1168 			return (error); /* NOTE: return error from nget */
1169 		}
1170 		newvp = NFSTOV(np);
1171 		if (lockparent) {
1172 			error = vn_lock(dvp, LK_EXCLUSIVE | LK_FAILRECLAIM);
1173 			if (error) {
1174 				vput(newvp);
1175 				lwkt_reltoken(&nmp->nm_token);
1176 				return (error);
1177 			}
1178 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1179 		}
1180 	} else if (NFS_CMPFH(np, fhp, fhsize)) {
1181 		vref(dvp);
1182 		newvp = dvp;
1183 	} else {
1184 		error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1185 		if (error) {
1186 			m_freem(info.mrep);
1187 			info.mrep = NULL;
1188 			lwkt_reltoken(&nmp->nm_token);
1189 			return (error);
1190 		}
1191 		if (!lockparent) {
1192 			vn_unlock(dvp);
1193 			cnp->cn_flags |= CNP_PDIRUNLOCK;
1194 		}
1195 		newvp = NFSTOV(np);
1196 	}
1197 	if (info.v3) {
1198 		ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1199 					  NFS_LATTR_NOSHRINK));
1200 		ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1201 					  NFS_LATTR_NOSHRINK));
1202 	} else {
1203 		ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1204 	}
1205 #if 0
1206 	/* XXX MOVE TO nfs_nremove() */
1207 	if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1208 	    cnp->cn_nameiop != NAMEI_DELETE) {
1209 		np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1210 	}
1211 #endif
1212 	*vpp = newvp;
1213 	m_freem(info.mrep);
1214 	info.mrep = NULL;
1215 nfsmout:
1216 	if (error) {
1217 		if (newvp != NULLVP) {
1218 			vrele(newvp);
1219 			*vpp = NULLVP;
1220 		}
1221 		if ((cnp->cn_nameiop == NAMEI_CREATE ||
1222 		     cnp->cn_nameiop == NAMEI_RENAME) &&
1223 		    error == ENOENT) {
1224 			if (!lockparent) {
1225 				vn_unlock(dvp);
1226 				cnp->cn_flags |= CNP_PDIRUNLOCK;
1227 			}
1228 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1229 				error = EROFS;
1230 			else
1231 				error = EJUSTRETURN;
1232 		}
1233 	}
1234 	lwkt_reltoken(&nmp->nm_token);
1235 	return (error);
1236 }
1237 
1238 /*
1239  * nfs read call.
1240  * Just call nfs_bioread() to do the work.
1241  *
1242  * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1243  *	    struct ucred *a_cred)
1244  */
1245 static int
1246 nfs_read(struct vop_read_args *ap)
1247 {
1248 	struct vnode *vp = ap->a_vp;
1249 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1250 	int error;
1251 
1252 	lwkt_gettoken(&nmp->nm_token);
1253 	error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag);
1254 	lwkt_reltoken(&nmp->nm_token);
1255 
1256 	return error;
1257 }
1258 
1259 /*
1260  * nfs readlink call
1261  *
1262  * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1263  */
1264 static int
1265 nfs_readlink(struct vop_readlink_args *ap)
1266 {
1267 	struct vnode *vp = ap->a_vp;
1268 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1269 	int error;
1270 
1271 	if (vp->v_type != VLNK)
1272 		return (EINVAL);
1273 
1274 	lwkt_gettoken(&nmp->nm_token);
1275 	error = nfs_bioread(vp, ap->a_uio, 0);
1276 	lwkt_reltoken(&nmp->nm_token);
1277 
1278 	return error;
1279 }
1280 
1281 /*
1282  * Do a readlink rpc.
1283  * Called by nfs_doio() from below the buffer cache.
1284  */
1285 int
1286 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1287 {
1288 	int error = 0, len, attrflag;
1289 	struct nfsm_info info;
1290 
1291 	info.mrep = NULL;
1292 	info.v3 = NFS_ISV3(vp);
1293 
1294 	nfsstats.rpccnt[NFSPROC_READLINK]++;
1295 	nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1296 	ERROROUT(nfsm_fhtom(&info, vp));
1297 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1298 				nfs_vpcred(vp, ND_CHECK), &error));
1299 	if (info.v3) {
1300 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1301 					  NFS_LATTR_NOSHRINK));
1302 	}
1303 	if (!error) {
1304 		NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1305 		if (len == NFS_MAXPATHLEN) {
1306 			struct nfsnode *np = VTONFS(vp);
1307 			if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1308 				len = np->n_size;
1309 		}
1310 		ERROROUT(nfsm_mtouio(&info, uiop, len));
1311 	}
1312 	m_freem(info.mrep);
1313 	info.mrep = NULL;
1314 nfsmout:
1315 	return (error);
1316 }
1317 
1318 /*
1319  * nfs synchronous read rpc using UIO
1320  */
1321 int
1322 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1323 {
1324 	u_int32_t *tl;
1325 	struct nfsmount *nmp;
1326 	int error = 0, len, retlen, tsiz, eof, attrflag;
1327 	struct nfsm_info info;
1328 	off_t tmp_off;
1329 
1330 	info.mrep = NULL;
1331 	info.v3 = NFS_ISV3(vp);
1332 
1333 #ifndef nolint
1334 	eof = 0;
1335 #endif
1336 	nmp = VFSTONFS(vp->v_mount);
1337 
1338 	tsiz = uiop->uio_resid;
1339 	tmp_off = uiop->uio_offset + tsiz;
1340 	if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1341 		return (EFBIG);
1342 	tmp_off = uiop->uio_offset;
1343 	while (tsiz > 0) {
1344 		nfsstats.rpccnt[NFSPROC_READ]++;
1345 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1346 		nfsm_reqhead(&info, vp, NFSPROC_READ,
1347 			     NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1348 		ERROROUT(nfsm_fhtom(&info, vp));
1349 		tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1350 		if (info.v3) {
1351 			txdr_hyper(uiop->uio_offset, tl);
1352 			*(tl + 2) = txdr_unsigned(len);
1353 		} else {
1354 			*tl++ = txdr_unsigned(uiop->uio_offset);
1355 			*tl++ = txdr_unsigned(len);
1356 			*tl = 0;
1357 		}
1358 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1359 					nfs_vpcred(vp, ND_READ), &error));
1360 		if (info.v3) {
1361 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1362 						 NFS_LATTR_NOSHRINK));
1363 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1364 			eof = fxdr_unsigned(int, *(tl + 1));
1365 		} else {
1366 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1367 		}
1368 		NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1369 		ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1370 		m_freem(info.mrep);
1371 		info.mrep = NULL;
1372 
1373 		/*
1374 		 * Handle short-read from server (NFSv3).  If EOF is not
1375 		 * flagged (and no error occurred), but retlen is less
1376 		 * then the request size, we must zero-fill the remainder.
1377 		 */
1378 		if (retlen < len && info.v3 && eof == 0) {
1379 			ERROROUT(uiomovez(len - retlen, uiop));
1380 			retlen = len;
1381 		}
1382 		tsiz -= retlen;
1383 
1384 		/*
1385 		 * Terminate loop on EOF or zero-length read.
1386 		 *
1387 		 * For NFSv2 a short-read indicates EOF, not zero-fill,
1388 		 * and also terminates the loop.
1389 		 */
1390 		if (info.v3) {
1391 			if (eof || retlen == 0)
1392 				tsiz = 0;
1393 		} else if (retlen < len) {
1394 			tsiz = 0;
1395 		}
1396 	}
1397 nfsmout:
1398 	return (error);
1399 }
1400 
1401 /*
1402  * nfs write call
1403  */
1404 int
1405 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1406 		 int *iomode, int *must_commit)
1407 {
1408 	u_int32_t *tl;
1409 	int32_t backup;
1410 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1411 	int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1412 	int  committed = NFSV3WRITE_FILESYNC;
1413 	struct nfsm_info info;
1414 
1415 	info.mrep = NULL;
1416 	info.v3 = NFS_ISV3(vp);
1417 
1418 #ifndef DIAGNOSTIC
1419 	if (uiop->uio_iovcnt != 1)
1420 		panic("nfs: writerpc iovcnt > 1");
1421 #endif
1422 	*must_commit = 0;
1423 	tsiz = uiop->uio_resid;
1424 	if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1425 		return (EFBIG);
1426 	while (tsiz > 0) {
1427 		nfsstats.rpccnt[NFSPROC_WRITE]++;
1428 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1429 		nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1430 			     NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED +
1431 			     nfsm_rndup(len));
1432 		ERROROUT(nfsm_fhtom(&info, vp));
1433 		if (info.v3) {
1434 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1435 			txdr_hyper(uiop->uio_offset, tl);
1436 			tl += 2;
1437 			*tl++ = txdr_unsigned(len);
1438 			*tl++ = txdr_unsigned(*iomode);
1439 			*tl = txdr_unsigned(len);
1440 		} else {
1441 			u_int32_t x;
1442 
1443 			tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1444 			/* Set both "begin" and "current" to non-garbage. */
1445 			x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1446 			*tl++ = x;	/* "begin offset" */
1447 			*tl++ = x;	/* "current offset" */
1448 			x = txdr_unsigned(len);
1449 			*tl++ = x;	/* total to this offset */
1450 			*tl = x;	/* size of this write */
1451 		}
1452 		ERROROUT(nfsm_uiotom(&info, uiop, len));
1453 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1454 					nfs_vpcred(vp, ND_WRITE), &error));
1455 		if (info.v3) {
1456 			/*
1457 			 * The write RPC returns a before and after mtime.  The
1458 			 * nfsm_wcc_data() macro checks the before n_mtime
1459 			 * against the before time and stores the after time
1460 			 * in the nfsnode's cached vattr and n_mtime field.
1461 			 * The NRMODIFIED bit will be set if the before
1462 			 * time did not match the original mtime.
1463 			 */
1464 			wccflag = NFSV3_WCCCHK;
1465 			ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1466 			if (error == 0) {
1467 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1468 				rlen = fxdr_unsigned(int, *tl++);
1469 				if (rlen == 0) {
1470 					error = NFSERR_IO;
1471 					m_freem(info.mrep);
1472 					info.mrep = NULL;
1473 					break;
1474 				} else if (rlen < len) {
1475 					backup = len - rlen;
1476 					uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1477 					uiop->uio_iov->iov_len += backup;
1478 					uiop->uio_offset -= backup;
1479 					uiop->uio_resid += backup;
1480 					len = rlen;
1481 				}
1482 				commit = fxdr_unsigned(int, *tl++);
1483 
1484 				/*
1485 				 * Return the lowest committment level
1486 				 * obtained by any of the RPCs.
1487 				 */
1488 				if (committed == NFSV3WRITE_FILESYNC)
1489 					committed = commit;
1490 				else if (committed == NFSV3WRITE_DATASYNC &&
1491 					commit == NFSV3WRITE_UNSTABLE)
1492 					committed = commit;
1493 				if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1494 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1495 					NFSX_V3WRITEVERF);
1496 				    nmp->nm_state |= NFSSTA_HASWRITEVERF;
1497 				} else if (bcmp((caddr_t)tl,
1498 				    (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1499 				    *must_commit = 1;
1500 				    bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1501 					NFSX_V3WRITEVERF);
1502 				}
1503 			}
1504 		} else {
1505 			ERROROUT(nfsm_loadattr(&info, vp, NULL));
1506 		}
1507 		m_freem(info.mrep);
1508 		info.mrep = NULL;
1509 		if (error)
1510 			break;
1511 		tsiz -= len;
1512 	}
1513 nfsmout:
1514 	if (vp->v_mount->mnt_flag & MNT_ASYNC)
1515 		committed = NFSV3WRITE_FILESYNC;
1516 	*iomode = committed;
1517 	if (error)
1518 		uiop->uio_resid = tsiz;
1519 	return (error);
1520 }
1521 
1522 /*
1523  * nfs mknod rpc
1524  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1525  * mode set to specify the file type and the size field for rdev.
1526  */
1527 static int
1528 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1529 	     struct vattr *vap)
1530 {
1531 	struct nfsv2_sattr *sp;
1532 	u_int32_t *tl;
1533 	struct vnode *newvp = NULL;
1534 	struct nfsnode *np = NULL;
1535 	struct vattr vattr;
1536 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1537 	int rmajor, rminor;
1538 	struct nfsm_info info;
1539 
1540 	info.mrep = NULL;
1541 	info.v3 = NFS_ISV3(dvp);
1542 
1543 	if (vap->va_type == VCHR || vap->va_type == VBLK) {
1544 		rmajor = txdr_unsigned(vap->va_rmajor);
1545 		rminor = txdr_unsigned(vap->va_rminor);
1546 	} else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1547 		rmajor = nfs_xdrneg1;
1548 		rminor = nfs_xdrneg1;
1549 	} else {
1550 		return (EOPNOTSUPP);
1551 	}
1552 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1553 		return (error);
1554 	}
1555 	nfsstats.rpccnt[NFSPROC_MKNOD]++;
1556 	nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1557 		     NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1558 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1559 	ERROROUT(nfsm_fhtom(&info, dvp));
1560 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1561 			     NFS_MAXNAMLEN));
1562 	if (info.v3) {
1563 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1564 		*tl++ = vtonfsv3_type(vap->va_type);
1565 		nfsm_v3attrbuild(&info, vap, FALSE);
1566 		if (vap->va_type == VCHR || vap->va_type == VBLK) {
1567 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1568 			*tl++ = txdr_unsigned(vap->va_rmajor);
1569 			*tl = txdr_unsigned(vap->va_rminor);
1570 		}
1571 	} else {
1572 		sp = nfsm_build(&info, NFSX_V2SATTR);
1573 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1574 		sp->sa_uid = nfs_xdrneg1;
1575 		sp->sa_gid = nfs_xdrneg1;
1576 		sp->sa_size = makeudev(rmajor, rminor);
1577 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1578 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1579 	}
1580 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1581 				cnp->cn_cred, &error));
1582 	if (!error) {
1583 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1584 		if (!gotvp) {
1585 			if (newvp) {
1586 				vput(newvp);
1587 				newvp = NULL;
1588 			}
1589 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1590 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1591 			if (!error)
1592 				newvp = NFSTOV(np);
1593 		}
1594 	}
1595 	if (info.v3) {
1596 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1597 	}
1598 	m_freem(info.mrep);
1599 	info.mrep = NULL;
1600 nfsmout:
1601 	if (error) {
1602 		if (newvp)
1603 			vput(newvp);
1604 	} else {
1605 		*vpp = newvp;
1606 	}
1607 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1608 	if (!wccflag)
1609 		VTONFS(dvp)->n_attrstamp = 0;
1610 	return (error);
1611 }
1612 
1613 /*
1614  * nfs mknod vop
1615  * just call nfs_mknodrpc() to do the work.
1616  *
1617  * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1618  *	     struct componentname *a_cnp, struct vattr *a_vap)
1619  */
1620 /* ARGSUSED */
1621 static int
1622 nfs_mknod(struct vop_old_mknod_args *ap)
1623 {
1624 	struct nfsmount *nmp = VFSTONFS(ap->a_dvp->v_mount);
1625 	int error;
1626 
1627 	lwkt_gettoken(&nmp->nm_token);
1628 	error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1629 	lwkt_reltoken(&nmp->nm_token);
1630 
1631 	return error;
1632 }
1633 
1634 static u_long create_verf;
1635 /*
1636  * nfs file create call
1637  *
1638  * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1639  *	      struct componentname *a_cnp, struct vattr *a_vap)
1640  */
1641 static int
1642 nfs_create(struct vop_old_create_args *ap)
1643 {
1644 	struct vnode *dvp = ap->a_dvp;
1645 	struct vattr *vap = ap->a_vap;
1646 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1647 	struct componentname *cnp = ap->a_cnp;
1648 	struct nfsv2_sattr *sp;
1649 	u_int32_t *tl;
1650 	struct nfsnode *np = NULL;
1651 	struct vnode *newvp = NULL;
1652 	int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1653 	struct vattr vattr;
1654 	struct nfsm_info info;
1655 
1656 	info.mrep = NULL;
1657 	info.v3 = NFS_ISV3(dvp);
1658 	lwkt_gettoken(&nmp->nm_token);
1659 
1660 	/*
1661 	 * Oops, not for me..
1662 	 */
1663 	if (vap->va_type == VSOCK) {
1664 		error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1665 		lwkt_reltoken(&nmp->nm_token);
1666 		return error;
1667 	}
1668 
1669 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1670 		lwkt_reltoken(&nmp->nm_token);
1671 		return (error);
1672 	}
1673 	if (vap->va_vaflags & VA_EXCLUSIVE)
1674 		fmode |= O_EXCL;
1675 again:
1676 	nfsstats.rpccnt[NFSPROC_CREATE]++;
1677 	nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1678 		     NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1679 		     nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1680 	ERROROUT(nfsm_fhtom(&info, dvp));
1681 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1682 			     NFS_MAXNAMLEN));
1683 	if (info.v3) {
1684 		tl = nfsm_build(&info, NFSX_UNSIGNED);
1685 		if (fmode & O_EXCL) {
1686 			*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1687 			tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1688 #ifdef INET
1689 			if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1690 				*tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1691 			else
1692 #endif
1693 				*tl++ = create_verf;
1694 			*tl = ++create_verf;
1695 		} else {
1696 			*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1697 			nfsm_v3attrbuild(&info, vap, FALSE);
1698 		}
1699 	} else {
1700 		sp = nfsm_build(&info, NFSX_V2SATTR);
1701 		sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1702 		sp->sa_uid = nfs_xdrneg1;
1703 		sp->sa_gid = nfs_xdrneg1;
1704 		sp->sa_size = 0;
1705 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1706 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1707 	}
1708 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1709 				cnp->cn_cred, &error));
1710 	if (error == 0) {
1711 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1712 		if (!gotvp) {
1713 			if (newvp) {
1714 				vput(newvp);
1715 				newvp = NULL;
1716 			}
1717 			error = nfs_lookitup(dvp, cnp->cn_nameptr,
1718 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1719 			if (!error)
1720 				newvp = NFSTOV(np);
1721 		}
1722 	}
1723 	if (info.v3) {
1724 		if (error == 0)
1725 			error = nfsm_wcc_data(&info, dvp, &wccflag);
1726 		else
1727 			(void)nfsm_wcc_data(&info, dvp, &wccflag);
1728 	}
1729 	m_freem(info.mrep);
1730 	info.mrep = NULL;
1731 nfsmout:
1732 	if (error) {
1733 		if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1734 			KKASSERT(newvp == NULL);
1735 			fmode &= ~O_EXCL;
1736 			goto again;
1737 		}
1738 	} else if (info.v3 && (fmode & O_EXCL)) {
1739 		/*
1740 		 * We are normally called with only a partially initialized
1741 		 * VAP.  Since the NFSv3 spec says that server may use the
1742 		 * file attributes to store the verifier, the spec requires
1743 		 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1744 		 * in atime, but we can't really assume that all servers will
1745 		 * so we ensure that our SETATTR sets both atime and mtime.
1746 		 */
1747 		if (vap->va_mtime.tv_sec == VNOVAL)
1748 			vfs_timestamp(&vap->va_mtime);
1749 		if (vap->va_atime.tv_sec == VNOVAL)
1750 			vap->va_atime = vap->va_mtime;
1751 		error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1752 	}
1753 	if (error == 0) {
1754 		/*
1755 		 * The new np may have enough info for access
1756 		 * checks, make sure rucred and wucred are
1757 		 * initialized for read and write rpc's.
1758 		 */
1759 		np = VTONFS(newvp);
1760 		if (np->n_rucred == NULL)
1761 			np->n_rucred = crhold(cnp->cn_cred);
1762 		if (np->n_wucred == NULL)
1763 			np->n_wucred = crhold(cnp->cn_cred);
1764 		*ap->a_vpp = newvp;
1765 	} else if (newvp) {
1766 		vput(newvp);
1767 	}
1768 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1769 	if (!wccflag)
1770 		VTONFS(dvp)->n_attrstamp = 0;
1771 	lwkt_reltoken(&nmp->nm_token);
1772 	return (error);
1773 }
1774 
1775 /*
1776  * nfs file remove call
1777  * To try and make nfs semantics closer to ufs semantics, a file that has
1778  * other processes using the vnode is renamed instead of removed and then
1779  * removed later on the last close.
1780  * - If v_refcnt > 1
1781  *	  If a rename is not already in the works
1782  *	     call nfs_sillyrename() to set it up
1783  *     else
1784  *	  do the remove rpc
1785  *
1786  * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1787  *	      struct componentname *a_cnp)
1788  */
1789 static int
1790 nfs_remove(struct vop_old_remove_args *ap)
1791 {
1792 	struct vnode *vp = ap->a_vp;
1793 	struct vnode *dvp = ap->a_dvp;
1794 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1795 	struct componentname *cnp = ap->a_cnp;
1796 	struct nfsnode *np = VTONFS(vp);
1797 	int error = 0;
1798 	struct vattr vattr;
1799 
1800 	lwkt_gettoken(&nmp->nm_token);
1801 #ifndef DIAGNOSTIC
1802 	if (VREFCNT(vp) < 1)
1803 		panic("nfs_remove: bad v_refcnt");
1804 #endif
1805 	if (vp->v_type == VDIR) {
1806 		error = EPERM;
1807 	} else if (VREFCNT(vp) == 1 || (np->n_sillyrename &&
1808 		   VOP_GETATTR(vp, &vattr) == 0 && vattr.va_nlink > 1)) {
1809 		/*
1810 		 * throw away biocache buffers, mainly to avoid
1811 		 * unnecessary delayed writes later.
1812 		 */
1813 		error = nfs_vinvalbuf(vp, 0, 1);
1814 		/* Do the rpc */
1815 		if (error != EINTR)
1816 			error = nfs_removerpc(dvp, cnp->cn_nameptr,
1817 				cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1818 		/*
1819 		 * Kludge City: If the first reply to the remove rpc is lost..
1820 		 *   the reply to the retransmitted request will be ENOENT
1821 		 *   since the file was in fact removed
1822 		 *   Therefore, we cheat and return success.
1823 		 */
1824 		if (error == ENOENT)
1825 			error = 0;
1826 	} else if (!np->n_sillyrename) {
1827 		error = nfs_sillyrename(dvp, vp, cnp);
1828 	}
1829 	np->n_attrstamp = 0;
1830 	lwkt_reltoken(&nmp->nm_token);
1831 
1832 	return (error);
1833 }
1834 
1835 /*
1836  * nfs file remove rpc called from nfs_inactive
1837  */
1838 int
1839 nfs_removeit(struct sillyrename *sp)
1840 {
1841 	return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1842 		sp->s_cred, NULL));
1843 }
1844 
1845 /*
1846  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1847  */
1848 static int
1849 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1850 	      struct ucred *cred, struct thread *td)
1851 {
1852 	int error = 0, wccflag = NFSV3_WCCRATTR;
1853 	struct nfsm_info info;
1854 
1855 	info.mrep = NULL;
1856 	info.v3 = NFS_ISV3(dvp);
1857 
1858 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
1859 	nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1860 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1861 	ERROROUT(nfsm_fhtom(&info, dvp));
1862 	ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1863 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1864 	if (info.v3) {
1865 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1866 	}
1867 	m_freem(info.mrep);
1868 	info.mrep = NULL;
1869 nfsmout:
1870 	VTONFS(dvp)->n_flag |= NLMODIFIED;
1871 	if (!wccflag)
1872 		VTONFS(dvp)->n_attrstamp = 0;
1873 	return (error);
1874 }
1875 
1876 /*
1877  * nfs file rename call
1878  *
1879  * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1880  *	      struct componentname *a_fcnp, struct vnode *a_tdvp,
1881  *	      struct vnode *a_tvp, struct componentname *a_tcnp)
1882  */
1883 static int
1884 nfs_rename(struct vop_old_rename_args *ap)
1885 {
1886 	struct vnode *fvp = ap->a_fvp;
1887 	struct vnode *tvp = ap->a_tvp;
1888 	struct vnode *fdvp = ap->a_fdvp;
1889 	struct vnode *tdvp = ap->a_tdvp;
1890 	struct componentname *tcnp = ap->a_tcnp;
1891 	struct componentname *fcnp = ap->a_fcnp;
1892 	struct nfsmount *nmp = VFSTONFS(fdvp->v_mount);
1893 	int error;
1894 
1895 	lwkt_gettoken(&nmp->nm_token);
1896 
1897 	/* Check for cross-device rename */
1898 	if ((fvp->v_mount != tdvp->v_mount) ||
1899 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1900 		error = EXDEV;
1901 		goto out;
1902 	}
1903 
1904 	/*
1905 	 * We shouldn't have to flush fvp on rename for most server-side
1906 	 * filesystems as the file handle should not change.  Unfortunately
1907 	 * the inode for some filesystems (msdosfs) might be tied to the
1908 	 * file name or directory position so to be completely safe
1909 	 * vfs.nfs.flush_on_rename is set by default.  Clear to improve
1910 	 * performance.
1911 	 *
1912 	 * We must flush tvp on rename because it might become stale on the
1913 	 * server after the rename.
1914 	 */
1915 	if (nfs_flush_on_rename)
1916 	    VOP_FSYNC(fvp, MNT_WAIT, 0);
1917 	if (tvp)
1918 	    VOP_FSYNC(tvp, MNT_WAIT, 0);
1919 
1920 	/*
1921 	 * If the tvp exists and is in use, sillyrename it before doing the
1922 	 * rename of the new file over it.
1923 	 *
1924 	 * XXX Can't sillyrename a directory.
1925 	 *
1926 	 * We do not attempt to do any namecache purges in this old API
1927 	 * routine.  The new API compat functions have access to the actual
1928 	 * namecache structures and will do it for us.
1929 	 */
1930 	if (tvp && VREFCNT(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1931 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1932 		vput(tvp);
1933 		tvp = NULL;
1934 	} else if (tvp) {
1935 		;
1936 	}
1937 
1938 	error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1939 		tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1940 		tcnp->cn_td);
1941 
1942 out:
1943 	lwkt_reltoken(&nmp->nm_token);
1944 	if (tdvp == tvp)
1945 		vrele(tdvp);
1946 	else
1947 		vput(tdvp);
1948 	if (tvp)
1949 		vput(tvp);
1950 	vrele(fdvp);
1951 	vrele(fvp);
1952 	/*
1953 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1954 	 */
1955 	if (error == ENOENT)
1956 		error = 0;
1957 	return (error);
1958 }
1959 
1960 /*
1961  * nfs file rename rpc called from nfs_remove() above
1962  */
1963 static int
1964 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1965 	     struct sillyrename *sp)
1966 {
1967 	return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1968 		sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1969 }
1970 
1971 /*
1972  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1973  */
1974 static int
1975 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1976 	      struct vnode *tdvp, const char *tnameptr, int tnamelen,
1977 	      struct ucred *cred, struct thread *td)
1978 {
1979 	int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1980 	struct nfsm_info info;
1981 
1982 	info.mrep = NULL;
1983 	info.v3 = NFS_ISV3(fdvp);
1984 
1985 	nfsstats.rpccnt[NFSPROC_RENAME]++;
1986 	nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1987 		    (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1988 		    nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1989 	ERROROUT(nfsm_fhtom(&info, fdvp));
1990 	ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1991 	ERROROUT(nfsm_fhtom(&info, tdvp));
1992 	ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1993 	NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1994 	if (info.v3) {
1995 		ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1996 		ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1997 	}
1998 	m_freem(info.mrep);
1999 	info.mrep = NULL;
2000 nfsmout:
2001 	VTONFS(fdvp)->n_flag |= NLMODIFIED;
2002 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2003 	if (!fwccflag)
2004 		VTONFS(fdvp)->n_attrstamp = 0;
2005 	if (!twccflag)
2006 		VTONFS(tdvp)->n_attrstamp = 0;
2007 	return (error);
2008 }
2009 
2010 /*
2011  * nfs hard link create call
2012  *
2013  * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
2014  *	    struct componentname *a_cnp)
2015  */
2016 static int
2017 nfs_link(struct vop_old_link_args *ap)
2018 {
2019 	struct vnode *vp = ap->a_vp;
2020 	struct vnode *tdvp = ap->a_tdvp;
2021 	struct nfsmount *nmp = VFSTONFS(tdvp->v_mount);
2022 	struct componentname *cnp = ap->a_cnp;
2023 	int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2024 	struct nfsm_info info;
2025 
2026 	if (vp->v_mount != tdvp->v_mount) {
2027 		return (EXDEV);
2028 	}
2029 	lwkt_gettoken(&nmp->nm_token);
2030 
2031 	/*
2032 	 * The attribute cache may get out of sync with the server on link.
2033 	 * Pushing writes to the server before handle was inherited from
2034 	 * long long ago and it is unclear if we still need to do this.
2035 	 * Defaults to off.
2036 	 */
2037 	if (nfs_flush_on_hlink)
2038 		VOP_FSYNC(vp, MNT_WAIT, 0);
2039 
2040 	info.mrep = NULL;
2041 	info.v3 = NFS_ISV3(vp);
2042 
2043 	nfsstats.rpccnt[NFSPROC_LINK]++;
2044 	nfsm_reqhead(&info, vp, NFSPROC_LINK,
2045 		     NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
2046 		     nfsm_rndup(cnp->cn_namelen));
2047 	ERROROUT(nfsm_fhtom(&info, vp));
2048 	ERROROUT(nfsm_fhtom(&info, tdvp));
2049 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2050 			     NFS_MAXNAMLEN));
2051 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
2052 				cnp->cn_cred, &error));
2053 	if (info.v3) {
2054 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2055 					 NFS_LATTR_NOSHRINK));
2056 		ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
2057 	}
2058 	m_freem(info.mrep);
2059 	info.mrep = NULL;
2060 nfsmout:
2061 	VTONFS(tdvp)->n_flag |= NLMODIFIED;
2062 	if (!attrflag)
2063 		VTONFS(vp)->n_attrstamp = 0;
2064 	if (!wccflag)
2065 		VTONFS(tdvp)->n_attrstamp = 0;
2066 	/*
2067 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2068 	 */
2069 	if (error == EEXIST)
2070 		error = 0;
2071 	lwkt_reltoken(&nmp->nm_token);
2072 	return (error);
2073 }
2074 
2075 /*
2076  * nfs symbolic link create call
2077  *
2078  * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
2079  *		struct componentname *a_cnp, struct vattr *a_vap,
2080  *		char *a_target)
2081  */
2082 static int
2083 nfs_symlink(struct vop_old_symlink_args *ap)
2084 {
2085 	struct vnode *dvp = ap->a_dvp;
2086 	struct vattr *vap = ap->a_vap;
2087 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2088 	struct componentname *cnp = ap->a_cnp;
2089 	struct nfsv2_sattr *sp;
2090 	int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2091 	struct vnode *newvp = NULL;
2092 	struct nfsm_info info;
2093 
2094 	info.mrep = NULL;
2095 	info.v3 = NFS_ISV3(dvp);
2096 	lwkt_gettoken(&nmp->nm_token);
2097 
2098 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2099 	slen = strlen(ap->a_target);
2100 	nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2101 		     NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2102 		     nfsm_rndup(cnp->cn_namelen) +
2103 		     nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2104 	ERROROUT(nfsm_fhtom(&info, dvp));
2105 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2106 			     NFS_MAXNAMLEN));
2107 	if (info.v3) {
2108 		nfsm_v3attrbuild(&info, vap, FALSE);
2109 	}
2110 	ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2111 	if (info.v3 == 0) {
2112 		sp = nfsm_build(&info, NFSX_V2SATTR);
2113 		sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2114 		sp->sa_uid = nfs_xdrneg1;
2115 		sp->sa_gid = nfs_xdrneg1;
2116 		sp->sa_size = nfs_xdrneg1;
2117 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2118 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2119 	}
2120 
2121 	/*
2122 	 * Issue the NFS request and get the rpc response.
2123 	 *
2124 	 * Only NFSv3 responses returning an error of 0 actually return
2125 	 * a file handle that can be converted into newvp without having
2126 	 * to do an extra lookup rpc.
2127 	 */
2128 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2129 				cnp->cn_cred, &error));
2130 	if (info.v3) {
2131 		if (error == 0) {
2132 		       ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2133 		}
2134 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2135 	}
2136 
2137 	/*
2138 	 * out code jumps -> here, mrep is also freed.
2139 	 */
2140 
2141 	m_freem(info.mrep);
2142 	info.mrep = NULL;
2143 nfsmout:
2144 
2145 	/*
2146 	 * If we get an EEXIST error, silently convert it to no-error
2147 	 * in case of an NFS retry.
2148 	 */
2149 	if (error == EEXIST)
2150 		error = 0;
2151 
2152 	/*
2153 	 * If we do not have (or no longer have) an error, and we could
2154 	 * not extract the newvp from the response due to the request being
2155 	 * NFSv2 or the error being EEXIST.  We have to do a lookup in order
2156 	 * to obtain a newvp to return.
2157 	 */
2158 	if (error == 0 && newvp == NULL) {
2159 		struct nfsnode *np = NULL;
2160 
2161 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2162 				     cnp->cn_cred, cnp->cn_td, &np);
2163 		if (!error)
2164 			newvp = NFSTOV(np);
2165 	}
2166 	if (error) {
2167 		if (newvp)
2168 			vput(newvp);
2169 	} else {
2170 		*ap->a_vpp = newvp;
2171 	}
2172 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2173 	if (!wccflag)
2174 		VTONFS(dvp)->n_attrstamp = 0;
2175 	lwkt_reltoken(&nmp->nm_token);
2176 
2177 	return (error);
2178 }
2179 
2180 /*
2181  * nfs make dir call
2182  *
2183  * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2184  *	     struct componentname *a_cnp, struct vattr *a_vap)
2185  */
2186 static int
2187 nfs_mkdir(struct vop_old_mkdir_args *ap)
2188 {
2189 	struct vnode *dvp = ap->a_dvp;
2190 	struct vattr *vap = ap->a_vap;
2191 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2192 	struct componentname *cnp = ap->a_cnp;
2193 	struct nfsv2_sattr *sp;
2194 	struct nfsnode *np = NULL;
2195 	struct vnode *newvp = NULL;
2196 	struct vattr vattr;
2197 	int error = 0, wccflag = NFSV3_WCCRATTR;
2198 	int gotvp = 0;
2199 	int len;
2200 	struct nfsm_info info;
2201 
2202 	info.mrep = NULL;
2203 	info.v3 = NFS_ISV3(dvp);
2204 	lwkt_gettoken(&nmp->nm_token);
2205 
2206 	if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2207 		lwkt_reltoken(&nmp->nm_token);
2208 		return (error);
2209 	}
2210 	len = cnp->cn_namelen;
2211 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
2212 	nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2213 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2214 		     nfsm_rndup(len) + NFSX_SATTR(info.v3));
2215 	ERROROUT(nfsm_fhtom(&info, dvp));
2216 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2217 	if (info.v3) {
2218 		nfsm_v3attrbuild(&info, vap, FALSE);
2219 	} else {
2220 		sp = nfsm_build(&info, NFSX_V2SATTR);
2221 		sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2222 		sp->sa_uid = nfs_xdrneg1;
2223 		sp->sa_gid = nfs_xdrneg1;
2224 		sp->sa_size = nfs_xdrneg1;
2225 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2226 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2227 	}
2228 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2229 		    cnp->cn_cred, &error));
2230 	if (error == 0) {
2231 		ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2232 	}
2233 	if (info.v3) {
2234 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2235 	}
2236 	m_freem(info.mrep);
2237 	info.mrep = NULL;
2238 nfsmout:
2239 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2240 	if (!wccflag)
2241 		VTONFS(dvp)->n_attrstamp = 0;
2242 	/*
2243 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2244 	 * if we can succeed in looking up the directory.
2245 	 */
2246 	if (error == EEXIST || (!error && !gotvp)) {
2247 		if (newvp) {
2248 			vrele(newvp);
2249 			newvp = NULL;
2250 		}
2251 		error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2252 			cnp->cn_td, &np);
2253 		if (!error) {
2254 			newvp = NFSTOV(np);
2255 			if (newvp->v_type != VDIR)
2256 				error = EEXIST;
2257 		}
2258 	}
2259 	if (error) {
2260 		if (newvp)
2261 			vrele(newvp);
2262 	} else {
2263 		*ap->a_vpp = newvp;
2264 	}
2265 	lwkt_reltoken(&nmp->nm_token);
2266 	return (error);
2267 }
2268 
2269 /*
2270  * nfs remove directory call
2271  *
2272  * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2273  *	     struct componentname *a_cnp)
2274  */
2275 static int
2276 nfs_rmdir(struct vop_old_rmdir_args *ap)
2277 {
2278 	struct vnode *vp = ap->a_vp;
2279 	struct vnode *dvp = ap->a_dvp;
2280 	struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2281 	struct componentname *cnp = ap->a_cnp;
2282 	int error = 0, wccflag = NFSV3_WCCRATTR;
2283 	struct nfsm_info info;
2284 
2285 	info.mrep = NULL;
2286 	info.v3 = NFS_ISV3(dvp);
2287 
2288 	if (dvp == vp)
2289 		return (EINVAL);
2290 
2291 	lwkt_gettoken(&nmp->nm_token);
2292 
2293 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
2294 	nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2295 		     NFSX_FH(info.v3) + NFSX_UNSIGNED +
2296 		     nfsm_rndup(cnp->cn_namelen));
2297 	ERROROUT(nfsm_fhtom(&info, dvp));
2298 	ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2299 		 NFS_MAXNAMLEN));
2300 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2301 				cnp->cn_cred, &error));
2302 	if (info.v3) {
2303 		ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2304 	}
2305 	m_freem(info.mrep);
2306 	info.mrep = NULL;
2307 nfsmout:
2308 	VTONFS(dvp)->n_flag |= NLMODIFIED;
2309 	if (!wccflag)
2310 		VTONFS(dvp)->n_attrstamp = 0;
2311 	/*
2312 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2313 	 */
2314 	if (error == ENOENT)
2315 		error = 0;
2316 	lwkt_reltoken(&nmp->nm_token);
2317 
2318 	return (error);
2319 }
2320 
2321 /*
2322  * nfs readdir call
2323  *
2324  * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2325  */
2326 static int
2327 nfs_readdir(struct vop_readdir_args *ap)
2328 {
2329 	struct vnode *vp = ap->a_vp;
2330 	struct nfsnode *np = VTONFS(vp);
2331 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2332 	struct uio *uio = ap->a_uio;
2333 	int tresid, error;
2334 	struct vattr vattr;
2335 
2336 	if (vp->v_type != VDIR)
2337 		return (EPERM);
2338 
2339 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
2340 	if (error)
2341 		return (error);
2342 
2343 	lwkt_gettoken(&nmp->nm_token);
2344 
2345 	/*
2346 	 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2347 	 * and then check that is still valid, or if this is an NQNFS mount
2348 	 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR().  Note that
2349 	 * VOP_GETATTR() does not necessarily go to the wire.
2350 	 */
2351 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2352 	    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2353 		if (VOP_GETATTR(vp, &vattr) == 0 &&
2354 		    (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2355 		) {
2356 			nfsstats.direofcache_hits++;
2357 			goto done;
2358 		}
2359 	}
2360 
2361 	/*
2362 	 * Call nfs_bioread() to do the real work.  nfs_bioread() does its
2363 	 * own cache coherency checks so we do not have to.
2364 	 */
2365 	tresid = uio->uio_resid;
2366 	error = nfs_bioread(vp, uio, 0);
2367 
2368 	if (!error && uio->uio_resid == tresid)
2369 		nfsstats.direofcache_misses++;
2370 done:
2371 	lwkt_reltoken(&nmp->nm_token);
2372 	vn_unlock(vp);
2373 
2374 	return (error);
2375 }
2376 
2377 /*
2378  * Readdir rpc call.  nfs_bioread->nfs_doio->nfs_readdirrpc.
2379  *
2380  * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2381  * offset/block and converts the nfs formatted directory entries for userland
2382  * consumption as well as deals with offsets into the middle of blocks.
2383  * nfs_doio only deals with logical blocks.  In particular, uio_offset will
2384  * be block-bounded.  It must convert to cookies for the actual RPC.
2385  */
2386 int
2387 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2388 {
2389 	int len, left;
2390 	struct nfs_dirent *dp = NULL;
2391 	u_int32_t *tl;
2392 	nfsuint64 *cookiep;
2393 	caddr_t cp;
2394 	nfsuint64 cookie;
2395 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2396 	struct nfsnode *dnp = VTONFS(vp);
2397 	u_quad_t fileno;
2398 	int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2399 	int attrflag;
2400 	struct nfsm_info info;
2401 
2402 	info.mrep = NULL;
2403 	info.v3 = NFS_ISV3(vp);
2404 
2405 #ifndef DIAGNOSTIC
2406 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2407 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2408 		panic("nfs readdirrpc bad uio");
2409 #endif
2410 
2411 	/*
2412 	 * If there is no cookie, assume directory was stale.
2413 	 */
2414 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2415 	if (cookiep)
2416 		cookie = *cookiep;
2417 	else
2418 		return (NFSERR_BAD_COOKIE);
2419 	/*
2420 	 * Loop around doing readdir rpc's of size nm_readdirsize
2421 	 * truncated to a multiple of DIRBLKSIZ.
2422 	 * The stopping criteria is EOF or buffer full.
2423 	 */
2424 	while (more_dirs && bigenough) {
2425 		nfsstats.rpccnt[NFSPROC_READDIR]++;
2426 		nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2427 			     NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2428 		ERROROUT(nfsm_fhtom(&info, vp));
2429 		if (info.v3) {
2430 			tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2431 			*tl++ = cookie.nfsuquad[0];
2432 			*tl++ = cookie.nfsuquad[1];
2433 			*tl++ = dnp->n_cookieverf.nfsuquad[0];
2434 			*tl++ = dnp->n_cookieverf.nfsuquad[1];
2435 		} else {
2436 			/*
2437 			 * WARNING!  HAMMER DIRECTORIES WILL NOT WORK WELL
2438 			 * WITH NFSv2!!!  There's nothing I can really do
2439 			 * about it other than to hope the server supports
2440 			 * rdirplus w/NFSv2.
2441 			 */
2442 			tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2443 			*tl++ = cookie.nfsuquad[0];
2444 		}
2445 		*tl = txdr_unsigned(nmp->nm_readdirsize);
2446 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2447 					uiop->uio_td,
2448 					nfs_vpcred(vp, ND_READ), &error));
2449 		if (info.v3) {
2450 			ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2451 						  NFS_LATTR_NOSHRINK));
2452 			NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2453 			dnp->n_cookieverf.nfsuquad[0] = *tl++;
2454 			dnp->n_cookieverf.nfsuquad[1] = *tl;
2455 		}
2456 		NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2457 		more_dirs = fxdr_unsigned(int, *tl);
2458 
2459 		/* loop thru the dir entries, converting them to std form */
2460 		while (more_dirs && bigenough) {
2461 			if (info.v3) {
2462 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2463 				fileno = fxdr_hyper(tl);
2464 				len = fxdr_unsigned(int, *(tl + 2));
2465 			} else {
2466 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2467 				fileno = fxdr_unsigned(u_quad_t, *tl++);
2468 				len = fxdr_unsigned(int, *tl);
2469 			}
2470 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2471 				error = EBADRPC;
2472 				m_freem(info.mrep);
2473 				info.mrep = NULL;
2474 				goto nfsmout;
2475 			}
2476 
2477 			/*
2478 			 * len is the number of bytes in the path element
2479 			 * name, not including the \0 termination.
2480 			 *
2481 			 * tlen is the number of bytes w have to reserve for
2482 			 * the path element name.
2483 			 */
2484 			tlen = nfsm_rndup(len);
2485 			if (tlen == len)
2486 				tlen += 4;	/* To ensure null termination */
2487 
2488 			/*
2489 			 * If the entry would cross a DIRBLKSIZ boundary,
2490 			 * extend the previous nfs_dirent to cover the
2491 			 * remaining space.
2492 			 */
2493 			left = DIRBLKSIZ - blksiz;
2494 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2495 				dp->nfs_reclen += left;
2496 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2497 				uiop->uio_iov->iov_len -= left;
2498 				uiop->uio_offset += left;
2499 				uiop->uio_resid -= left;
2500 				blksiz = 0;
2501 			}
2502 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2503 				bigenough = 0;
2504 			if (bigenough) {
2505 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2506 				dp->nfs_ino = fileno;
2507 				dp->nfs_namlen = len;
2508 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2509 				dp->nfs_type = DT_UNKNOWN;
2510 				blksiz += dp->nfs_reclen;
2511 				if (blksiz == DIRBLKSIZ)
2512 					blksiz = 0;
2513 				uiop->uio_offset += sizeof(struct nfs_dirent);
2514 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2515 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2516 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2517 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2518 
2519 				/*
2520 				 * The uiop has advanced by nfs_dirent + len
2521 				 * but really needs to advance by
2522 				 * nfs_dirent + tlen
2523 				 */
2524 				cp = uiop->uio_iov->iov_base;
2525 				tlen -= len;
2526 				*cp = '\0';	/* null terminate */
2527 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2528 				uiop->uio_iov->iov_len -= tlen;
2529 				uiop->uio_offset += tlen;
2530 				uiop->uio_resid -= tlen;
2531 			} else {
2532 				/*
2533 				 * NFS strings must be rounded up (nfsm_myouio
2534 				 * handled that in the bigenough case).
2535 				 */
2536 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2537 			}
2538 			if (info.v3) {
2539 				NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2540 			} else {
2541 				NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2542 			}
2543 
2544 			/*
2545 			 * If we were able to accomodate the last entry,
2546 			 * get the cookie for the next one.  Otherwise
2547 			 * hold-over the cookie for the one we were not
2548 			 * able to accomodate.
2549 			 */
2550 			if (bigenough) {
2551 				cookie.nfsuquad[0] = *tl++;
2552 				if (info.v3)
2553 					cookie.nfsuquad[1] = *tl++;
2554 			} else if (info.v3) {
2555 				tl += 2;
2556 			} else {
2557 				tl++;
2558 			}
2559 			more_dirs = fxdr_unsigned(int, *tl);
2560 		}
2561 		/*
2562 		 * If at end of rpc data, get the eof boolean
2563 		 */
2564 		if (!more_dirs) {
2565 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2566 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2567 		}
2568 		m_freem(info.mrep);
2569 		info.mrep = NULL;
2570 	}
2571 	/*
2572 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2573 	 * by increasing d_reclen for the last record.
2574 	 */
2575 	if (blksiz > 0) {
2576 		left = DIRBLKSIZ - blksiz;
2577 		dp->nfs_reclen += left;
2578 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2579 		uiop->uio_iov->iov_len -= left;
2580 		uiop->uio_offset += left;
2581 		uiop->uio_resid -= left;
2582 	}
2583 
2584 	if (bigenough) {
2585 		/*
2586 		 * We hit the end of the directory, update direofoffset.
2587 		 */
2588 		dnp->n_direofoffset = uiop->uio_offset;
2589 	} else {
2590 		/*
2591 		 * There is more to go, insert the link cookie so the
2592 		 * next block can be read.
2593 		 */
2594 		if (uiop->uio_resid > 0)
2595 			kprintf("EEK! readdirrpc resid > 0\n");
2596 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2597 		*cookiep = cookie;
2598 	}
2599 nfsmout:
2600 	return (error);
2601 }
2602 
2603 /*
2604  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2605  */
2606 int
2607 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2608 {
2609 	int len, left;
2610 	struct nfs_dirent *dp;
2611 	u_int32_t *tl;
2612 	struct vnode *newvp;
2613 	nfsuint64 *cookiep;
2614 	caddr_t dpossav1, dpossav2;
2615 	caddr_t cp;
2616 	struct mbuf *mdsav1, *mdsav2;
2617 	nfsuint64 cookie;
2618 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2619 	struct nfsnode *dnp = VTONFS(vp), *np;
2620 	nfsfh_t *fhp;
2621 	u_quad_t fileno;
2622 	int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2623 	int attrflag, fhsize;
2624 	struct nchandle nch;
2625 	struct nchandle dnch;
2626 	struct nlcomponent nlc;
2627 	struct nfsm_info info;
2628 
2629 	info.mrep = NULL;
2630 	info.v3 = 1;
2631 
2632 #ifndef nolint
2633 	dp = NULL;
2634 #endif
2635 #ifndef DIAGNOSTIC
2636 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2637 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2638 		panic("nfs readdirplusrpc bad uio");
2639 #endif
2640 	/*
2641 	 * Obtain the namecache record for the directory so we have something
2642 	 * to use as a basis for creating the entries.  This function will
2643 	 * return a held (but not locked) ncp.  The ncp may be disconnected
2644 	 * from the tree and cannot be used for upward traversals, and the
2645 	 * ncp may be unnamed.  Note that other unrelated operations may
2646 	 * cause the ncp to be named at any time.
2647 	 *
2648 	 * We have to lock the ncp to prevent a lock order reversal when
2649 	 * rdirplus does nlookups of the children, because the vnode is
2650 	 * locked and has to stay that way.
2651 	 */
2652 	cache_fromdvp(vp, NULL, 0, &dnch);
2653 	bzero(&nlc, sizeof(nlc));
2654 	newvp = NULLVP;
2655 
2656 	/*
2657 	 * If there is no cookie, assume directory was stale.
2658 	 */
2659 	cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2660 	if (cookiep) {
2661 		cookie = *cookiep;
2662 	} else {
2663 		if (dnch.ncp)
2664 			cache_drop(&dnch);
2665 		return (NFSERR_BAD_COOKIE);
2666 	}
2667 
2668 	/*
2669 	 * Loop around doing readdir rpc's of size nm_readdirsize
2670 	 * truncated to a multiple of DIRBLKSIZ.
2671 	 * The stopping criteria is EOF or buffer full.
2672 	 */
2673 	while (more_dirs && bigenough) {
2674 		nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2675 		nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2676 			     NFSX_FH(info.v3) + 6 * NFSX_UNSIGNED);
2677 		ERROROUT(nfsm_fhtom(&info, vp));
2678 		tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2679 		*tl++ = cookie.nfsuquad[0];
2680 		*tl++ = cookie.nfsuquad[1];
2681 		*tl++ = dnp->n_cookieverf.nfsuquad[0];
2682 		*tl++ = dnp->n_cookieverf.nfsuquad[1];
2683 		*tl++ = txdr_unsigned(nmp->nm_readdirsize);
2684 		*tl = txdr_unsigned(nmp->nm_rsize);
2685 		NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2686 					uiop->uio_td,
2687 					nfs_vpcred(vp, ND_READ), &error));
2688 		ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2689 					  NFS_LATTR_NOSHRINK));
2690 		NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2691 		dnp->n_cookieverf.nfsuquad[0] = *tl++;
2692 		dnp->n_cookieverf.nfsuquad[1] = *tl++;
2693 		more_dirs = fxdr_unsigned(int, *tl);
2694 
2695 		/* loop thru the dir entries, doctoring them to 4bsd form */
2696 		while (more_dirs && bigenough) {
2697 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2698 			fileno = fxdr_hyper(tl);
2699 			len = fxdr_unsigned(int, *(tl + 2));
2700 			if (len <= 0 || len > NFS_MAXNAMLEN) {
2701 				error = EBADRPC;
2702 				m_freem(info.mrep);
2703 				info.mrep = NULL;
2704 				goto nfsmout;
2705 			}
2706 			tlen = nfsm_rndup(len);
2707 			if (tlen == len)
2708 				tlen += 4;	/* To ensure null termination*/
2709 			left = DIRBLKSIZ - blksiz;
2710 			if ((tlen + sizeof(struct nfs_dirent)) > left) {
2711 				dp->nfs_reclen += left;
2712 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2713 				uiop->uio_iov->iov_len -= left;
2714 				uiop->uio_offset += left;
2715 				uiop->uio_resid -= left;
2716 				blksiz = 0;
2717 			}
2718 			if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2719 				bigenough = 0;
2720 			if (bigenough) {
2721 				dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2722 				dp->nfs_ino = fileno;
2723 				dp->nfs_namlen = len;
2724 				dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2725 				dp->nfs_type = DT_UNKNOWN;
2726 				blksiz += dp->nfs_reclen;
2727 				if (blksiz == DIRBLKSIZ)
2728 					blksiz = 0;
2729 				uiop->uio_offset += sizeof(struct nfs_dirent);
2730 				uiop->uio_resid -= sizeof(struct nfs_dirent);
2731 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2732 				uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2733 				nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2734 				nlc.nlc_namelen = len;
2735 				ERROROUT(nfsm_mtouio(&info, uiop, len));
2736 				cp = uiop->uio_iov->iov_base;
2737 				tlen -= len;
2738 				*cp = '\0';
2739 				uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2740 				uiop->uio_iov->iov_len -= tlen;
2741 				uiop->uio_offset += tlen;
2742 				uiop->uio_resid -= tlen;
2743 			} else {
2744 				ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2745 			}
2746 			NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2747 			if (bigenough) {
2748 				cookie.nfsuquad[0] = *tl++;
2749 				cookie.nfsuquad[1] = *tl++;
2750 			} else {
2751 				tl += 2;
2752 			}
2753 
2754 			/*
2755 			 * Since the attributes are before the file handle
2756 			 * (sigh), we must skip over the attributes and then
2757 			 * come back and get them.
2758 			 */
2759 			attrflag = fxdr_unsigned(int, *tl);
2760 			if (attrflag) {
2761 			    dpossav1 = info.dpos;
2762 			    mdsav1 = info.md;
2763 			    ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2764 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2765 			    doit = fxdr_unsigned(int, *tl);
2766 			    if (doit) {
2767 				NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2768 			    }
2769 			    if (doit && bigenough && !nlcdegenerate(&nlc) &&
2770 				!NFS_CMPFH(dnp, fhp, fhsize)
2771 			    ) {
2772 				if (dnch.ncp) {
2773 #if 0
2774 				    kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2775 					nlc.nlc_namelen, nlc.nlc_namelen,
2776 					nlc.nlc_nameptr);
2777 #endif
2778 				    /*
2779 				     * This is a bit hokey but there isn't
2780 				     * much we can do about it.  We can't
2781 				     * hold the directory vp locked while
2782 				     * doing lookups and gets.
2783 				     */
2784 				    nch = cache_nlookup_nonblock(&dnch, &nlc);
2785 				    if (nch.ncp == NULL)
2786 					goto rdfail;
2787 				    cache_setunresolved(&nch);
2788 				    error = nfs_nget_nonblock(vp->v_mount, fhp,
2789 							      fhsize, &np,
2790 							      NULL);
2791 				    if (error) {
2792 					cache_put(&nch);
2793 					goto rdfail;
2794 				    }
2795 				    newvp = NFSTOV(np);
2796 				    dpossav2 = info.dpos;
2797 				    info.dpos = dpossav1;
2798 				    mdsav2 = info.md;
2799 				    info.md = mdsav1;
2800 				    ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2801 				    info.dpos = dpossav2;
2802 				    info.md = mdsav2;
2803 				    dp->nfs_type =
2804 					    IFTODT(VTTOIF(np->n_vattr.va_type));
2805 				    nfs_cache_setvp(&nch, newvp,
2806 						    nfspos_cache_timeout);
2807 				    vput(newvp);
2808 				    newvp = NULLVP;
2809 				    cache_put(&nch);
2810 				} else {
2811 rdfail:
2812 				    ;
2813 #if 0
2814 				    kprintf("Warning: NFS/rddirplus, "
2815 					    "UNABLE TO ENTER %*.*s\n",
2816 					nlc.nlc_namelen, nlc.nlc_namelen,
2817 					nlc.nlc_nameptr);
2818 #endif
2819 				}
2820 			    }
2821 			} else {
2822 			    /* Just skip over the file handle */
2823 			    NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2824 			    i = fxdr_unsigned(int, *tl);
2825 			    ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2826 			}
2827 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2828 			more_dirs = fxdr_unsigned(int, *tl);
2829 		}
2830 		/*
2831 		 * If at end of rpc data, get the eof boolean
2832 		 */
2833 		if (!more_dirs) {
2834 			NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2835 			more_dirs = (fxdr_unsigned(int, *tl) == 0);
2836 		}
2837 		m_freem(info.mrep);
2838 		info.mrep = NULL;
2839 	}
2840 	/*
2841 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2842 	 * by increasing d_reclen for the last record.
2843 	 */
2844 	if (blksiz > 0) {
2845 		left = DIRBLKSIZ - blksiz;
2846 		dp->nfs_reclen += left;
2847 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2848 		uiop->uio_iov->iov_len -= left;
2849 		uiop->uio_offset += left;
2850 		uiop->uio_resid -= left;
2851 	}
2852 
2853 	/*
2854 	 * We are now either at the end of the directory or have filled the
2855 	 * block.
2856 	 */
2857 	if (bigenough) {
2858 		dnp->n_direofoffset = uiop->uio_offset;
2859 	} else {
2860 		if (uiop->uio_resid > 0)
2861 			kprintf("EEK! readdirplusrpc resid > 0\n");
2862 		cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2863 		*cookiep = cookie;
2864 	}
2865 nfsmout:
2866 	if (newvp != NULLVP) {
2867 	        if (newvp == vp)
2868 			vrele(newvp);
2869 		else
2870 			vput(newvp);
2871 		newvp = NULLVP;
2872 	}
2873 	if (dnch.ncp)
2874 		cache_drop(&dnch);
2875 	return (error);
2876 }
2877 
2878 /*
2879  * Silly rename. To make the NFS filesystem that is stateless look a little
2880  * more like the "ufs" a remove of an active vnode is translated to a rename
2881  * to a funny looking filename that is removed by nfs_inactive on the
2882  * nfsnode. There is the potential for another process on a different client
2883  * to create the same funny name between the nfs_lookitup() fails and the
2884  * nfs_rename() completes, but...
2885  */
2886 static int
2887 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2888 {
2889 	struct sillyrename *sp;
2890 	struct nfsnode *np;
2891 	int error;
2892 
2893 	/*
2894 	 * We previously purged dvp instead of vp.  I don't know why, it
2895 	 * completely destroys performance.  We can't do it anyway with the
2896 	 * new VFS API since we would be breaking the namecache topology.
2897 	 */
2898 	cache_purge(vp);	/* XXX */
2899 	np = VTONFS(vp);
2900 #ifndef DIAGNOSTIC
2901 	if (vp->v_type == VDIR)
2902 		panic("nfs: sillyrename dir");
2903 #endif
2904 	sp = kmalloc(sizeof(struct sillyrename), M_NFSREQ, M_WAITOK);
2905 	sp->s_cred = crdup(cnp->cn_cred);
2906 	sp->s_dvp = dvp;
2907 	vref(dvp);
2908 
2909 	/* Fudge together a funny name */
2910 	sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2911 				(int)(intptr_t)cnp->cn_td);
2912 
2913 	/* Try lookitups until we get one that isn't there */
2914 	while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2915 		cnp->cn_td, NULL) == 0) {
2916 		sp->s_name[4]++;
2917 		if (sp->s_name[4] > 'z') {
2918 			error = EINVAL;
2919 			goto bad;
2920 		}
2921 	}
2922 	error = nfs_renameit(dvp, cnp, sp);
2923 	if (error)
2924 		goto bad;
2925 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2926 		cnp->cn_td, &np);
2927 	np->n_sillyrename = sp;
2928 	return (0);
2929 bad:
2930 	vrele(sp->s_dvp);
2931 	crfree(sp->s_cred);
2932 	kfree((caddr_t)sp, M_NFSREQ);
2933 	return (error);
2934 }
2935 
2936 /*
2937  * Look up a file name and optionally either update the file handle or
2938  * allocate an nfsnode, depending on the value of npp.
2939  * npp == NULL	--> just do the lookup
2940  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2941  *			handled too
2942  * *npp != NULL --> update the file handle in the vnode
2943  */
2944 static int
2945 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2946 	     struct thread *td, struct nfsnode **npp)
2947 {
2948 	struct vnode *newvp = NULL;
2949 	struct nfsnode *np, *dnp = VTONFS(dvp);
2950 	int error = 0, fhlen, attrflag;
2951 	nfsfh_t *nfhp;
2952 	struct nfsm_info info;
2953 
2954 	info.mrep = NULL;
2955 	info.v3 = NFS_ISV3(dvp);
2956 
2957 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2958 	nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2959 		     NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2960 	ERROROUT(nfsm_fhtom(&info, dvp));
2961 	ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2962 	NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2963 	if (npp && !error) {
2964 		NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2965 		if (*npp) {
2966 		    np = *npp;
2967 		    if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2968 			kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2969 			np->n_fhp = &np->n_fh;
2970 		    } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2971 			np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2972 		    bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2973 		    np->n_fhsize = fhlen;
2974 		    newvp = NFSTOV(np);
2975 		} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2976 		    vref(dvp);
2977 		    newvp = dvp;
2978 		} else {
2979 		    error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, NULL);
2980 		    if (error) {
2981 			m_freem(info.mrep);
2982 			info.mrep = NULL;
2983 			return (error);
2984 		    }
2985 		    newvp = NFSTOV(np);
2986 		}
2987 		if (info.v3) {
2988 			ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2989 						  NFS_LATTR_NOSHRINK));
2990 			if (!attrflag && *npp == NULL) {
2991 				m_freem(info.mrep);
2992 				info.mrep = NULL;
2993 				if (newvp == dvp)
2994 					vrele(newvp);
2995 				else
2996 					vput(newvp);
2997 				return (ENOENT);
2998 			}
2999 		} else {
3000 			ERROROUT(nfsm_loadattr(&info, newvp, NULL));
3001 		}
3002 	}
3003 	m_freem(info.mrep);
3004 	info.mrep = NULL;
3005 nfsmout:
3006 	if (npp && *npp == NULL) {
3007 		if (error) {
3008 			if (newvp) {
3009 				if (newvp == dvp)
3010 					vrele(newvp);
3011 				else
3012 					vput(newvp);
3013 			}
3014 		} else
3015 			*npp = np;
3016 	}
3017 	return (error);
3018 }
3019 
3020 /*
3021  * Nfs Version 3 commit rpc
3022  *
3023  * We call it 'uio' to distinguish it from 'bio' but there is no real uio
3024  * involved.
3025  */
3026 int
3027 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
3028 {
3029 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3030 	int error = 0, wccflag = NFSV3_WCCRATTR;
3031 	struct nfsm_info info;
3032 	u_int32_t *tl;
3033 
3034 	info.mrep = NULL;
3035 	info.v3 = 1;
3036 
3037 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
3038 		return (0);
3039 	nfsstats.rpccnt[NFSPROC_COMMIT]++;
3040 	nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
3041 	ERROROUT(nfsm_fhtom(&info, vp));
3042 	tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
3043 	txdr_hyper(offset, tl);
3044 	tl += 2;
3045 	*tl = txdr_unsigned(cnt);
3046 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
3047 				nfs_vpcred(vp, ND_WRITE), &error));
3048 	ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
3049 	if (!error) {
3050 		NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
3051 		if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
3052 			NFSX_V3WRITEVERF)) {
3053 			bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
3054 				NFSX_V3WRITEVERF);
3055 			error = NFSERR_STALEWRITEVERF;
3056 		}
3057 	}
3058 	m_freem(info.mrep);
3059 	info.mrep = NULL;
3060 nfsmout:
3061 	return (error);
3062 }
3063 
3064 /*
3065  * Kludge City..
3066  * - make nfs_bmap() essentially a no-op that does no translation
3067  * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3068  *   (Maybe I could use the process's page mapping, but I was concerned that
3069  *    Kernel Write might not be enabled and also figured copyout() would do
3070  *    a lot more work than bcopy() and also it currently happens in the
3071  *    context of the swapper process (2).
3072  *
3073  * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
3074  *	    off_t *a_doffsetp, int *a_runp, int *a_runb)
3075  */
3076 static int
3077 nfs_bmap(struct vop_bmap_args *ap)
3078 {
3079 	/* no token lock required */
3080 	if (ap->a_doffsetp != NULL)
3081 		*ap->a_doffsetp = ap->a_loffset;
3082 	if (ap->a_runp != NULL)
3083 		*ap->a_runp = 0;
3084 	if (ap->a_runb != NULL)
3085 		*ap->a_runb = 0;
3086 	return (0);
3087 }
3088 
3089 /*
3090  * Strategy routine.
3091  */
3092 static int
3093 nfs_strategy(struct vop_strategy_args *ap)
3094 {
3095 	struct bio *bio = ap->a_bio;
3096 	struct bio *nbio;
3097 	struct buf *bp __debugvar = bio->bio_buf;
3098 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3099 	struct thread *td;
3100 	int error;
3101 
3102 	KASSERT(bp->b_cmd != BUF_CMD_DONE,
3103 		("nfs_strategy: buffer %p unexpectedly marked done", bp));
3104 	KASSERT(BUF_REFCNT(bp) > 0,
3105 		("nfs_strategy: buffer %p not locked", bp));
3106 
3107 	if (bio->bio_flags & BIO_SYNC)
3108 		td = curthread;	/* XXX */
3109 	else
3110 		td = NULL;
3111 
3112 	lwkt_gettoken(&nmp->nm_token);
3113 
3114         /*
3115 	 * We probably don't need to push an nbio any more since no
3116 	 * block conversion is required due to the use of 64 bit byte
3117 	 * offsets, but do it anyway.
3118 	 *
3119 	 * NOTE: When NFS callers itself via this strategy routines and
3120 	 *	 sets up a synchronous I/O, it expects the I/O to run
3121 	 *	 synchronously (its bio_done routine just assumes it),
3122 	 *	 so for now we have to honor the bit.
3123          */
3124 	nbio = push_bio(bio);
3125 	nbio->bio_offset = bio->bio_offset;
3126 	nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3127 
3128 	/*
3129 	 * If the op is asynchronous and an i/o daemon is waiting
3130 	 * queue the request, wake it up and wait for completion
3131 	 * otherwise just do it ourselves.
3132 	 */
3133 	if (bio->bio_flags & BIO_SYNC) {
3134 		error = nfs_doio(ap->a_vp, nbio, td);
3135 	} else {
3136 		nfs_asyncio(ap->a_vp, nbio);
3137 		error = 0;
3138 	}
3139 	lwkt_reltoken(&nmp->nm_token);
3140 
3141 	return (error);
3142 }
3143 
3144 /*
3145  * Mmap a file
3146  *
3147  * NB Currently unsupported.
3148  *
3149  * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3150  */
3151 /* ARGSUSED */
3152 static int
3153 nfs_mmap(struct vop_mmap_args *ap)
3154 {
3155 	/* no token lock required */
3156 	return (EINVAL);
3157 }
3158 
3159 /*
3160  * fsync vnode op. Just call nfs_flush() with commit == 1.
3161  *
3162  * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3163  */
3164 /* ARGSUSED */
3165 static int
3166 nfs_fsync(struct vop_fsync_args *ap)
3167 {
3168 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3169 	int error;
3170 
3171 	lwkt_gettoken(&nmp->nm_token);
3172 
3173 	/*
3174 	 * NOTE: Because attributes are set synchronously we currently
3175 	 *	 do not have to implement vsetisdirty()/vclrisdirty().
3176 	 */
3177 	error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1);
3178 
3179 	lwkt_reltoken(&nmp->nm_token);
3180 
3181 	return error;
3182 }
3183 
3184 /*
3185  * Flush all the blocks associated with a vnode.   Dirty NFS buffers may be
3186  * in one of two states:  If B_NEEDCOMMIT is clear then the buffer contains
3187  * new NFS data which needs to be written to the server.  If B_NEEDCOMMIT is
3188  * set the buffer contains data that has already been written to the server
3189  * and which now needs a commit RPC.
3190  *
3191  * If commit is 0 we only take one pass and only flush buffers containing new
3192  * dirty data.
3193  *
3194  * If commit is 1 we take two passes, issuing a commit RPC in the second
3195  * pass.
3196  *
3197  * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3198  * to completely flush all pending data.
3199  *
3200  * Note that the RB_SCAN code properly handles the case where the
3201  * callback might block and directly or indirectly (another thread) cause
3202  * the RB tree to change.
3203  */
3204 
3205 #ifndef NFS_COMMITBVECSIZ
3206 #define NFS_COMMITBVECSIZ	16
3207 #endif
3208 
3209 struct nfs_flush_info {
3210 	enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3211 	struct thread *td;
3212 	struct vnode *vp;
3213 	int waitfor;
3214 	int slpflag;
3215 	int slptimeo;
3216 	int loops;
3217 	struct buf *bvary[NFS_COMMITBVECSIZ];
3218 	int bvsize;
3219 	off_t beg_off;
3220 	off_t end_off;
3221 };
3222 
3223 static int nfs_flush_bp(struct buf *bp, void *data);
3224 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3225 
3226 int
3227 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3228 {
3229 	struct nfsnode *np = VTONFS(vp);
3230 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3231 	struct nfs_flush_info info;
3232 	int error;
3233 
3234 	bzero(&info, sizeof(info));
3235 	info.td = td;
3236 	info.vp = vp;
3237 	info.waitfor = waitfor;
3238 	info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3239 	info.loops = 0;
3240 	lwkt_gettoken(&vp->v_token);
3241 
3242 	do {
3243 		/*
3244 		 * Flush mode
3245 		 */
3246 		info.mode = NFI_FLUSHNEW;
3247 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3248 				nfs_flush_bp, &info);
3249 
3250 		/*
3251 		 * Take a second pass if committing and no error occured.
3252 		 * Clean up any left over collection (whether an error
3253 		 * occurs or not).
3254 		 */
3255 		if (commit && error == 0) {
3256 			info.mode = NFI_COMMIT;
3257 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3258 					nfs_flush_bp, &info);
3259 			if (info.bvsize)
3260 				error = nfs_flush_docommit(&info, error);
3261 		}
3262 
3263 		/*
3264 		 * Wait for pending I/O to complete before checking whether
3265 		 * any further dirty buffers exist.
3266 		 */
3267 		while (waitfor == MNT_WAIT &&
3268 		       bio_track_active(&vp->v_track_write)) {
3269 			error = bio_track_wait(&vp->v_track_write,
3270 					       info.slpflag, info.slptimeo);
3271 			if (error) {
3272 				/*
3273 				 * We have to be able to break out if this
3274 				 * is an 'intr' mount.
3275 				 */
3276 				if (nfs_sigintr(nmp, NULL, td)) {
3277 					error = -EINTR;
3278 					break;
3279 				}
3280 
3281 				/*
3282 				 * Since we do not process pending signals,
3283 				 * once we get a PCATCH our tsleep() will no
3284 				 * longer sleep, switch to a fixed timeout
3285 				 * instead.
3286 				 */
3287 				if (info.slpflag == PCATCH) {
3288 					info.slpflag = 0;
3289 					info.slptimeo = 2 * hz;
3290 				}
3291 				error = 0;
3292 			}
3293 		}
3294 		++info.loops;
3295 		/*
3296 		 * Loop if we are flushing synchronous as well as committing,
3297 		 * and dirty buffers are still present.  Otherwise we might livelock.
3298 		 */
3299 	} while (waitfor == MNT_WAIT && commit &&
3300 		 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3301 
3302 	/*
3303 	 * The callbacks have to return a negative error to terminate the
3304 	 * RB scan.
3305 	 */
3306 	if (error < 0)
3307 		error = -error;
3308 
3309 	/*
3310 	 * Deal with any error collection
3311 	 */
3312 	if (np->n_flag & NWRITEERR) {
3313 		error = np->n_error;
3314 		np->n_flag &= ~NWRITEERR;
3315 	}
3316 	lwkt_reltoken(&vp->v_token);
3317 	return (error);
3318 }
3319 
3320 static
3321 int
3322 nfs_flush_bp(struct buf *bp, void *data)
3323 {
3324 	struct nfs_flush_info *info = data;
3325 	int lkflags;
3326 	int error;
3327 	off_t toff;
3328 
3329 	error = 0;
3330 	switch(info->mode) {
3331 	case NFI_FLUSHNEW:
3332 		error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3333 		if (error && info->loops && info->waitfor == MNT_WAIT) {
3334 			error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3335 			if (error) {
3336 				lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3337 				if (info->slpflag & PCATCH)
3338 					lkflags |= LK_PCATCH;
3339 				error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3340 						     info->slptimeo);
3341 			}
3342 		}
3343 
3344 		/*
3345 		 * Ignore locking errors
3346 		 */
3347 		if (error) {
3348 			error = 0;
3349 			break;
3350 		}
3351 
3352 		/*
3353 		 * The buffer may have changed out from under us, even if
3354 		 * we did not block (MPSAFE).  Check again now that it is
3355 		 * locked.
3356 		 */
3357 		if (bp->b_vp == info->vp &&
3358 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3359 			bremfree(bp);
3360 			bawrite(bp);
3361 		} else {
3362 			BUF_UNLOCK(bp);
3363 		}
3364 		break;
3365 	case NFI_COMMIT:
3366 		/*
3367 		 * Only process buffers in need of a commit which we can
3368 		 * immediately lock.  This may prevent a buffer from being
3369 		 * committed, but the normal flush loop will block on the
3370 		 * same buffer so we shouldn't get into an endless loop.
3371 		 */
3372 		if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3373 		    (B_DELWRI | B_NEEDCOMMIT)) {
3374 			break;
3375 		}
3376 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3377 			break;
3378 
3379 		/*
3380 		 * We must recheck after successfully locking the buffer.
3381 		 */
3382 		if (bp->b_vp != info->vp ||
3383 		    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3384 		    (B_DELWRI | B_NEEDCOMMIT)) {
3385 			BUF_UNLOCK(bp);
3386 			break;
3387 		}
3388 
3389 		/*
3390 		 * NOTE: storing the bp in the bvary[] basically sets
3391 		 * it up for a commit operation.
3392 		 *
3393 		 * We must call vfs_busy_pages() now so the commit operation
3394 		 * is interlocked with user modifications to memory mapped
3395 		 * pages.  The b_dirtyoff/b_dirtyend range is not correct
3396 		 * until after the pages have been busied.
3397 		 *
3398 		 * Note: to avoid loopback deadlocks, we do not
3399 		 * assign b_runningbufspace.
3400 		 */
3401 		bremfree(bp);
3402 		bp->b_cmd = BUF_CMD_WRITE;
3403 		vfs_busy_pages(bp->b_vp, bp);
3404 		info->bvary[info->bvsize] = bp;
3405 		toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3406 		if (info->bvsize == 0 || toff < info->beg_off)
3407 			info->beg_off = toff;
3408 		toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3409 		if (info->bvsize == 0 || toff > info->end_off)
3410 			info->end_off = toff;
3411 		++info->bvsize;
3412 		if (info->bvsize == NFS_COMMITBVECSIZ) {
3413 			error = nfs_flush_docommit(info, 0);
3414 			KKASSERT(info->bvsize == 0);
3415 		}
3416 	}
3417 	return (error);
3418 }
3419 
3420 static
3421 int
3422 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3423 {
3424 	struct vnode *vp;
3425 	struct buf *bp;
3426 	off_t bytes;
3427 	int retv;
3428 	int i;
3429 
3430 	vp = info->vp;
3431 
3432 	if (info->bvsize > 0) {
3433 		/*
3434 		 * Commit data on the server, as required.  Note that
3435 		 * nfs_commit will use the vnode's cred for the commit.
3436 		 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3437 		 */
3438 		bytes = info->end_off - info->beg_off;
3439 		if (bytes > 0x40000000)
3440 			bytes = 0x40000000;
3441 		if (error) {
3442 			retv = -error;
3443 		} else {
3444 			retv = nfs_commitrpc_uio(vp, info->beg_off,
3445 						 (int)bytes, info->td);
3446 			if (retv == NFSERR_STALEWRITEVERF)
3447 				nfs_clearcommit(vp->v_mount);
3448 		}
3449 
3450 		/*
3451 		 * Now, either mark the blocks I/O done or mark the
3452 		 * blocks dirty, depending on whether the commit
3453 		 * succeeded.
3454 		 */
3455 		for (i = 0; i < info->bvsize; ++i) {
3456 			bp = info->bvary[i];
3457 			if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3458 				/*
3459 				 * Either an error or the original
3460 				 * vfs_busy_pages() cleared B_NEEDCOMMIT
3461 				 * due to finding new dirty VM pages in
3462 				 * the buffer.
3463 				 *
3464 				 * Leave B_DELWRI intact.
3465 				 */
3466 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3467 				vfs_unbusy_pages(bp);
3468 				bp->b_cmd = BUF_CMD_DONE;
3469 				bqrelse(bp);
3470 			} else {
3471 				/*
3472 				 * Success, remove B_DELWRI ( bundirty() ).
3473 				 *
3474 				 * b_dirtyoff/b_dirtyend seem to be NFS
3475 				 * specific.  We should probably move that
3476 				 * into bundirty(). XXX
3477 				 *
3478 				 * We are faking an I/O write, we have to
3479 				 * start the transaction in order to
3480 				 * immediately biodone() it.
3481 				 */
3482 				bundirty(bp);
3483 				bp->b_flags &= ~B_ERROR;
3484 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3485 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3486 				biodone(&bp->b_bio1);
3487 			}
3488 		}
3489 		info->bvsize = 0;
3490 	}
3491 	return (error);
3492 }
3493 
3494 /*
3495  * NFS advisory byte-level locks.
3496  * Currently unsupported.
3497  *
3498  * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3499  *		int a_flags)
3500  */
3501 static int
3502 nfs_advlock(struct vop_advlock_args *ap)
3503 {
3504 	struct nfsnode *np = VTONFS(ap->a_vp);
3505 
3506 	/* no token lock currently required */
3507 	/*
3508 	 * The following kludge is to allow diskless support to work
3509 	 * until a real NFS lockd is implemented. Basically, just pretend
3510 	 * that this is a local lock.
3511 	 */
3512 	return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3513 }
3514 
3515 /*
3516  * Print out the contents of an nfsnode.
3517  *
3518  * nfs_print(struct vnode *a_vp)
3519  */
3520 static int
3521 nfs_print(struct vop_print_args *ap)
3522 {
3523 	struct vnode *vp = ap->a_vp;
3524 	struct nfsnode *np = VTONFS(vp);
3525 
3526 	kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3527 		(long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3528 	if (vp->v_type == VFIFO)
3529 		fifo_printinfo(vp);
3530 	kprintf("\n");
3531 	return (0);
3532 }
3533 
3534 /*
3535  * nfs special file access vnode op.
3536  *
3537  * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3538  */
3539 static int
3540 nfs_laccess(struct vop_access_args *ap)
3541 {
3542 	struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3543 	struct vattr vattr;
3544 	int error;
3545 
3546 	lwkt_gettoken(&nmp->nm_token);
3547 	error = VOP_GETATTR(ap->a_vp, &vattr);
3548 	if (error == 0) {
3549 		error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3550 					  vattr.va_mode, 0);
3551 	}
3552 	lwkt_reltoken(&nmp->nm_token);
3553 
3554 	return (error);
3555 }
3556 
3557 /*
3558  * Read wrapper for fifos.
3559  *
3560  * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3561  *		struct ucred *a_cred)
3562  */
3563 static int
3564 nfsfifo_read(struct vop_read_args *ap)
3565 {
3566 	struct nfsnode *np = VTONFS(ap->a_vp);
3567 
3568 	/* no token access required */
3569 	/*
3570 	 * Set access flag.
3571 	 */
3572 	np->n_flag |= NACC;
3573 	getnanotime(&np->n_atim);
3574 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3575 }
3576 
3577 /*
3578  * Write wrapper for fifos.
3579  *
3580  * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3581  *		 struct ucred *a_cred)
3582  */
3583 static int
3584 nfsfifo_write(struct vop_write_args *ap)
3585 {
3586 	struct nfsnode *np = VTONFS(ap->a_vp);
3587 
3588 	/* no token access required */
3589 	/*
3590 	 * Set update flag.
3591 	 */
3592 	np->n_flag |= NUPD;
3593 	getnanotime(&np->n_mtim);
3594 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3595 }
3596 
3597 /*
3598  * Close wrapper for fifos.
3599  *
3600  * Update the times on the nfsnode then do fifo close.
3601  *
3602  * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3603  */
3604 static int
3605 nfsfifo_close(struct vop_close_args *ap)
3606 {
3607 	struct vnode *vp = ap->a_vp;
3608 	struct nfsnode *np = VTONFS(vp);
3609 	struct vattr vattr;
3610 	struct timespec ts;
3611 
3612 	/* no token access required */
3613 
3614 	vn_lock(vp, LK_UPGRADE | LK_RETRY); /* XXX */
3615 	if (np->n_flag & (NACC | NUPD)) {
3616 		getnanotime(&ts);
3617 		if (np->n_flag & NACC)
3618 			np->n_atim = ts;
3619 		if (np->n_flag & NUPD)
3620 			np->n_mtim = ts;
3621 		np->n_flag |= NCHG;
3622 		if (VREFCNT(vp) == 1 &&
3623 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3624 			VATTR_NULL(&vattr);
3625 			if (np->n_flag & NACC)
3626 				vattr.va_atime = np->n_atim;
3627 			if (np->n_flag & NUPD)
3628 				vattr.va_mtime = np->n_mtim;
3629 			(void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3630 		}
3631 	}
3632 	return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3633 }
3634 
3635